The Inference Report

Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process. However, most existing pipelines only utilize the language modality for scene descriptions or reasoning and lack the flexibility to follow diverse user instructions for personalized driving. To address this, we first construct a large-scale driving dataset (InstructScene) containing around 100,000 scenes annotated with diverse driving instructions with the corresponding trajectories. We then propose a unified Vision-Language-World-Action model, Vega, for instruction-based generation and planning. We employ the autoregressive paradigm to process visual inputs (vision) and language instructions (language) and the diffusion paradigm to generate future predictions (world modeling) and trajectories (action). We perform joint attention to enable interactions between the modalities and use individual projection layers for different modalities for more capabilities. Extensive experiments demonstrate that our method not only achieves superior planning performance but also exhibits strong instruction-following abilities, paving the way for more intelligent and personalized driving systems.

Human driving behavior is inherently personal, which is shaped by long-term habits and influenced by short-term intentions. Individuals differ in how they accelerate, brake, merge, yield, and overtake across diverse situations. However, existing end-to-end autonomous driving systems either optimize for generic objectives or rely on fixed driving modes, lacking the ability to adapt to individual preferences or interpret natural language intent. To address this gap, we propose Drive My Way (DMW), a personalized Vision-Language-Action (VLA) driving framework that aligns with users' long-term driving habits and adapts to real-time user instructions. DMW learns a user embedding from our personalized driving dataset collected across multiple real drivers and conditions the policy on this embedding during planning, while natural language instructions provide additional short-term guidance. Closed-loop evaluation on the Bench2Drive benchmark demonstrates that DMW improves style instruction adaptation, and user studies show that its generated behaviors are recognizable as each driver's own style, highlighting personalization as a key capability for human-centered autonomous driving. Our data and code are available at https://dmw-cvpr.github.io/.

The knowledge base in a retrieval-augmented generation (RAG) system is typically assembled once and never revised, even though the facts a query requires are often fragmented across documents and buried in irrelevant content. We argue that the knowledge base should be treated as a trainable component and propose WriteBack-RAG, a framework that uses labeled examples to identify where retrieval succeeds, isolate the relevant documents, and distill them into compact knowledge units that are indexed alongside the original corpus. Because the method modifies only the corpus, it can be applied once as an offline preprocessing step and combined with any RAG pipeline. Across four RAG methods, six benchmarks, and two LLM backbones, WriteBack-RAG improves every evaluated setting, with gains averaging +2.14%. Cross-method transfer experiments further show that the distilled knowledge benefits RAG pipelines other than the one used to produce it, confirming that the improvement resides in the corpus itself.

Autoregressive video diffusion models have demonstrated remarkable progress, yet they remain bottlenecked by intractable linear KV-cache growth, temporal repetition, and compounding errors during long-video generation. To address these challenges, we present PackForcing, a unified framework that efficiently manages the generation history through a novel three-partition KV-cache strategy. Specifically, we categorize the historical context into three distinct types: (1) Sink tokens, which preserve early anchor frames at full resolution to maintain global semantics; (2) Mid tokens, which achieve a massive spatiotemporal compression (32x token reduction) via a dual-branch network fusing progressive 3D convolutions with low-resolution VAE re-encoding; and (3) Recent tokens, kept at full resolution to ensure local temporal coherence. To strictly bound the memory footprint without sacrificing quality, we introduce a dynamic top-$k$ context selection mechanism for the mid tokens, coupled with a continuous Temporal RoPE Adjustment that seamlessly re-aligns position gaps caused by dropped tokens with negligible overhead. Empowered by this principled hierarchical context compression, PackForcing can generate coherent 2-minute, 832x480 videos at 16 FPS on a single H200 GPU. It achieves a bounded KV cache of just 4 GB and enables a remarkable 24x temporal extrapolation (5s to 120s), operating effectively either zero-shot or trained on merely 5-second clips. Extensive results on VBench demonstrate state-of-the-art temporal consistency (26.07) and dynamic degree (56.25), proving that short-video supervision is sufficient for high-quality, long-video synthesis. https://github.com/ShandaAI/PackForcing

Fine-grained facial expression editing has long been limited by intrinsic semantic overlap. To address this, we construct the Flex Facial Expression (FFE) dataset with continuous affective annotations and establish FFE-Bench to evaluate structural confusion, editing accuracy, linear controllability, and the trade-off between expression editing and identity preservation. We propose PixelSmile, a diffusion framework that disentangles expression semantics via fully symmetric joint training. PixelSmile combines intensity supervision with contrastive learning to produce stronger and more distinguishable expressions, achieving precise and stable linear expression control through textual latent interpolation. Extensive experiments demonstrate that PixelSmile achieves superior disentanglement and robust identity preservation, confirming its effectiveness for continuous, controllable, and fine-grained expression editing, while naturally supporting smooth expression blending.

Automatic speech recognition (ASR) systems have achieved near-human accuracy on curated benchmarks, yet still fail in real-world voice agents under conditions that current evaluations do not systematically cover. Without diagnostic tools that isolate specific failure factors, practitioners cannot anticipate which conditions, in which languages, will cause what degree of degradation. We introduce WildASR, a multilingual (four-language) diagnostic benchmark sourced entirely from real human speech that factorizes ASR robustness along three axes: environmental degradation, demographic shift, and linguistic diversity. Evaluating seven widely used ASR systems, we find severe and uneven performance degradation, and model robustness does not transfer across languages or conditions. Critically, models often hallucinate plausible but unspoken content under partial or degraded inputs, creating concrete safety risks for downstream agent behavior. Our results demonstrate that targeted, factor-isolated evaluation is essential for understanding and improving ASR reliability in production systems. Besides the benchmark itself, we also present three analytical tools that practitioners can use to guide deployment decisions.

Agent performance increasingly depends on \emph{harness engineering}, yet harness design is usually buried in controller code and runtime-specific conventions, making it hard to transfer, compare, and study as a scientific object. We ask whether the high-level control logic of an agent harness can instead be externalized as a portable executable artifact. We introduce \textbf{Natural-Language Agent Harnesses} (NLAHs), which express harness behavior in editable natural language, and \textbf{Intelligent Harness Runtime} (IHR), a shared runtime that executes these harnesses through explicit contracts, durable artifacts, and lightweight adapters. Across coding and computer-use benchmarks, we conduct controlled evaluations of operational viability, module ablation, and code-to-text harness migration.

Contrastive vision-language (V&L) models remain a popular choice for various applications. However, several limitations have emerged, most notably the limited ability of V&L models to learn compositional representations. Prior methods often addressed this limitation by generating custom training data to obtain hard negative samples. Hard negatives have been shown to improve performance on compositionality tasks, but are often specific to a single benchmark, do not generalize, and can cause substantial degradation of basic V&L capabilities such as zero-shot or retrieval performance, rendering them impractical. In this work we follow a different approach. We identify two root causes that limit compositionality performance of V&Ls: 1) Long training captions do not require a compositional representation; and 2) The final global pooling in the text and image encoders lead to a complete loss of the necessary information to learn binding in the first place. As a remedy, we propose two simple solutions: 1) We obtain short concept centric caption parts using standard NLP software and align those with the image; and 2) We introduce a parameter-free cross-modal attention-pooling to obtain concept centric visual embeddings from the image encoder. With these two changes and simple auxiliary contrastive losses, we obtain SOTA performance on standard compositionality benchmarks, while maintaining or improving strong zero-shot and retrieval capabilities. This is achieved without increasing inference cost. We release the code for this work at https://github.com/SamsungLabs/concept_centric_clip.

Robust perception and reasoning require consistency across sensory modalities. Yet current multimodal models often violate this principle, yielding contradictory predictions for visual and textual representations of the same concept. Rather than masking these failures with standard voting mechanisms, which can amplify systematic biases, we show that cross-modal inconsistency provides a rich and natural signal for learning. We introduce RC2, a reinforcement learning framework that resolves internal conflicts by enforcing cross-modal cycle consistency. By requiring a model to perform backward inference, switch modalities, and reliably reconstruct the answer through forward inference, we obtain a dense, label-free reward. This cyclic constraint encourages the model to align its internal representations autonomously. Optimizing for this structure mitigates modality-specific errors and improves reasoning accuracy by up to 7.6 points. Our results suggest that advanced reasoning emerges not only from scaling data, but also from enforcing a structurally consistent understanding of the world.

We present an empirical study of how far general-purpose coding agents -- without hardware-specific training -- can optimize hardware designs from high-level algorithmic specifications. We introduce an agent factory, a two-stage pipeline that constructs and coordinates multiple autonomous optimization agents. In Stage~1, the pipeline decomposes a design into sub-kernels, independently optimizes each using pragma and code-level transformations, and formulates an Integer Linear Program (ILP) to assemble globally promising configurations under an area constraint. In Stage~2, it launches $N$ expert agents over the top ILP solutions, each exploring cross-function optimizations such as pragma recombination, loop fusion, and memory restructuring that are not captured by sub-kernel decomposition. We evaluate the approach on 12 kernels from HLS-Eval and Rodinia-HLS using Claude Code (Opus~4.5/4.6) with AMD Vitis HLS. Scaling from 1 to 10 agents yields a mean $8.27\times$ speedup over baseline, with larger gains on harder benchmarks: streamcluster exceeds $20\times$ and kmeans reaches approximately $10\times$. Across benchmarks, agents consistently rediscover known hardware optimization patterns without domain-specific training, and the best designs often do not originate from top-ranked ILP candidates, indicating that global optimization exposes improvements missed by sub-kernel search. These results establish agent scaling as a practical and effective axis for HLS optimization.

Video world models have shown immense potential in simulating the physical world, yet existing memory mechanisms primarily treat environments as static canvases. When dynamic subjects hide out of sight and later re-emerge, current methods often struggle, leading to frozen, distorted, or vanishing subjects. To address this, we introduce Hybrid Memory, a novel paradigm requiring models to simultaneously act as precise archivists for static backgrounds and vigilant trackers for dynamic subjects, ensuring motion continuity during out-of-view intervals. To facilitate research in this direction, we construct HM-World, the first large-scale video dataset dedicated to hybrid memory. It features 59K high-fidelity clips with decoupled camera and subject trajectories, encompassing 17 diverse scenes, 49 distinct subjects, and meticulously designed exit-entry events to rigorously evaluate hybrid coherence. Furthermore, we propose HyDRA, a specialized memory architecture that compresses memory into tokens and utilizes a spatiotemporal relevance-driven retrieval mechanism. By selectively attending to relevant motion cues, HyDRA effectively preserves the identity and motion of hidden subjects. Extensive experiments on HM-World demonstrate that our method significantly outperforms state-of-the-art approaches in both dynamic subject consistency and overall generation quality.

Block-diffusion language models offer a promising path toward faster-than-autoregressive generation by combining block-wise autoregressive decoding with within-block parallel denoising. However, in the few-step regime needed for practical acceleration, standard confidence-thresholded decoding is often brittle: aggressive thresholds hurt quality, while conservative thresholds require unnecessary denoising steps. Existing approaches that address this issue either require additional training or incur extra test-time compute. We present S2D2, a training-free self-speculative decoding framework for block-diffusion language models. Our key observation is that a block-diffusion model becomes autoregressive when the block size is reduced to one, allowing the same pretrained model to act as both drafter and verifier. S2D2 inserts a speculative verification step into standard block-diffusion decoding and uses lightweight routing policies to decide when verification is worth its cost. This yields a hybrid decoding trajectory in which diffusion proposes tokens in parallel, while the autoregressive mode acts as a local sequence-level critic. Across three mainstream block-diffusion families, S2D2 consistently improves the accuracy-speed tradeoff over strong confidence-thresholding baselines. On SDAR, we observe up to $4.7\times$ speedup over autoregressive decoding, and up to $1.57\times$ over a tuned dynamic decoding baseline while improving accuracy by up to $4.5$ points. On LLaDA2.1-Mini, S2D2 remains complementary to built-in self-correction, including a conservative setting where it is $4.4\times$ faster than the static baseline with slightly higher accuracy.

Transfer learning and knowledge distillation has recently gained a lot of attention in the deep learning community. One transfer approach, the student-teacher learning, has been shown to successfully create ``small'' student neural networks that mimic the performance of a much bigger and more complex ``teacher'' networks. In this paper, we investigate an extension to this approach and transfer from a non-neural-based machine learning pipeline as teacher to a neural network (NN) student, which would allow for joint optimization of the various pipeline components and a single unified inference engine for multiple ML tasks. In particular, we explore replacing the random forest classifier by transfer learning to a student NN. We experimented with various NN topologies on 100 OpenML tasks in which random forest has been one of the best solutions. Our results show that for the majority of the tasks, the student NN can indeed mimic the teacher if one can select the right NN hyper-parameters. We also investigated the use of random forest for selecting the right NN hyper-parameters.

Code production is now a commodity; the bottleneck is knowing what to build and proving it works. We present the Kitchen Loop, a framework for autonomous, self-evolving software built on a unified trust model: (1) a specification surface enumerating what the product claims to support; (2) 'As a User x 1000', where an LLM agent exercises that surface as a synthetic power user at 1,000x human cadence; (3) Unbeatable Tests, ground-truth verification the code author cannot fake; and (4) Drift Control, continuous quality measurement with automated pause gates. We validate across two production systems over 285+ iterations, producing 1,094+ merged pull requests with zero regressions detected by the regression oracle (methodology in Section 6.1). We observe emergent properties at scale: multi-iteration self-correction chains, autonomous infrastructure healing, and monotonically improving quality gates. The primitives are not new; our contribution is their composition into a production-tested system with the operational discipline that makes long-running autonomous evolution safe.

Trustworthy artificial intelligence increasingly relies on probabilistic computation to achieve robustness, interpretability, security and privacy. In practical systems, such workloads interleave deterministic data access with repeated stochastic sampling across models, data paths and system functions, shifting performance bottlenecks from arithmetic units to memory systems that must deliver both data and randomness. Here we present a unified data-access perspective in which deterministic access is treated as a limiting case of stochastic sampling, enabling both modes to be analyzed within a common framework. This view reveals that increasing stochastic demand reduces effective data-access efficiency and can drive systems into entropy-limited operation. Based on this insight, we define memory-level evaluation criteria, including unified operation, distribution programmability, efficiency, robustness to hardware non-idealities and parallel compatibility. Using these criteria, we analyze limitations of conventional architectures and examine emerging probabilistic compute-in-memory approaches that integrate sampling with memory access, outlining pathways toward scalable hardware for trustworthy AI.

Neural scaling laws, which in some domains can predict the performance of large neural networks as a function of model, data, and compute scale, are the cornerstone of building foundation models in Natural Language Processing and Computer Vision. We study neural scaling in Scientific Machine Learning, focusing on models for weather forecasting. To analyze scaling behavior in as simple a setting as possible, we adopt a minimal, scalable, general-purpose Swin Transformer architecture, and we use continual training with constant learning rates and periodic cooldowns as an efficient training strategy. We show that models trained in this minimalist way follow predictable scaling trends and even outperform standard cosine learning rate schedules. Cooldown phases can be re-purposed to improve downstream performance, e.g., enabling accurate multi-step rollouts over longer forecast horizons as well as sharper predictions through spectral loss adjustments. We also systematically explore a wide range of model and dataset sizes under various compute budgets to construct IsoFLOP curves, and we identify compute-optimal training regimes. Extrapolating these trends to larger scales highlights potential performance limits, demonstrating that neural scaling can serve as an important diagnostic for efficient resource allocation. We open-source our code for reproducibility.

Cross-view geo-localization (CVGL) estimates a camera's location by matching a street-view image to geo-referenced overhead imagery, enabling GPS-denied localization and navigation. Existing methods almost universally formulate CVGL as an image-retrieval problem in a contrastively trained embedding space. This ties performance to large batches and hard negative mining, and it ignores both the geometric structure of maps and the coverage mismatch between street-view and overhead imagery. In particular, salient landmarks visible from the street view can fall outside a fixed satellite crop, making retrieval targets ambiguous and limiting explicit spatial inference over the map. We propose Just Zoom In, an alternative formulation that performs CVGL via autoregressive zooming over a city-scale overhead map. Starting from a coarse satellite view, the model takes a short sequence of zoom-in decisions to select a terminal satellite cell at a target resolution, without contrastive losses or hard negative mining. We further introduce a realistic benchmark with crowd-sourced street views and high-resolution satellite imagery that reflects real capture conditions. On this benchmark, Just Zoom In achieves state-of-the-art performance, improving Recall@1 within 50 m by 5.5% and Recall@1 within 100 m by 9.6% over the strongest contrastive-retrieval baseline. These results demonstrate the effectiveness of sequential coarse-to-fine spatial reasoning for cross-view geo-localization.

As large language models (LLMs) continue to advance, improving them solely through human supervision is becoming increasingly costly and limited in scalability. As models approach human-level capabilities in certain domains, human feedback may no longer provide sufficiently informative signals for further improvement. At the same time, the growing ability of models to make autonomous decisions and execute complex actions naturally enables abstractions in which components of the model development process can be progressively automated. Together, these challenges and opportunities have driven increasing interest in self-improvement, where models autonomously generate data, evaluate outputs, and iteratively refine their own capabilities. In this paper, we present a system-level perspective on self-improving language models and introduce a unified framework that organizes existing techniques. We conceptualize the self-improvement system as a closed-loop lifecycle, consisting of four tightly coupled processes: data acquisition, data selection, model optimization, and inference refinement, along with an autonomous evaluation layer. Within this framework, the model itself plays a central role in driving each stage: collecting or generating data, selecting informative signals, updating its parameters, and refining outputs, while the autonomous evaluation layer continuously monitors progress and guides the improvement cycle across stages. Following this lifecycle perspective, we systematically review and analyze representative methods for each component from a technical standpoint. We further discuss current limitations and outline our vision for future research toward fully self-improving LLMs.

Automated systems have been widely adopted across the educational testing industry for open-response assessment and essay scoring. These systems commonly achieve performance levels comparable to or superior than trained human raters, but have frequently been demonstrated to be vulnerable to the influence of construct-irrelevant factors (i.e., features of responses that are unrelated to the construct assessed) and adversarial conditions. Given the rising usage of large language models in automated scoring systems, there is a renewed focus on ``hallucinations'' and the robustness of these LLM-based automated scoring approaches to construct-irrelevant factors. This study investigates the effects of construct-irrelevant factors on a dual-architecture LLM-based scoring system designed to score short essay-like open-response items in a situational judgment test. It was found that the scoring system was generally robust to padding responses with meaningless text, spelling errors, and writing sophistication. Duplicating large passages of text resulted in lower scores predicted by the system, on average, contradicting results from previous studies of non-LLM-based scoring systems, while off-topic responses were heavily penalized by the scoring system. These results provide encouraging support for the robustness of future LLM-based scoring systems when designed with construct relevance in mind.

Early detection of atypical cognitive-motor development is critical for timely intervention, yet traditional assessments rely heavily on subjective, static evaluations. The integration of digital devices offers an opportunity for continuous, objective monitoring through digital biomarkers. In this work, we propose an AI-driven longitudinal framework to model developmental trajectories in children aged 18 months to 8 years. Using a dataset of tablet-based interactions collected over multiple academic years, we analyzed six cognitive-motor tasks (e.g., fine motor control, reaction time). We applied dimensionality reduction (t-SNE) and unsupervised clustering (K-Means++) to identify distinct developmental phenotypes and tracked individual transitions between these profiles over time. Our analysis reveals three distinct profiles: low, medium, and high performance. Crucially, longitudinal tracking highlights a high stability in the low-performance cluster (>90% retention in early years), suggesting that early deficits tend to persist without intervention. Conversely, higher-performance clusters show greater variability, potentially reflecting engagement factors. This study validates the use of unsupervised learning on touchscreen data to uncover heterogeneous developmental paths. The identified profiles serve as scalable, data-driven proxies for cognitive growth, offering a foundation for early screening tools and personalized pediatric interventions.

Safety monitoring is essential for Cyber-Physical Systems (CPSs). However, unsafe events are rare in real-world CPS operations, creating an extreme class imbalance that degrades safety predictors. Standard rebalancing techniques perform poorly on time-series CPS telemetry, either generating unrealistic synthetic samples or overfitting on the minority class. Meanwhile, behavioral uncertainty in CPS operations, defined as the degree of doubt or uncertainty in CPS decisions , is often correlated with safety outcomes but unexplored in safety monitoring. To that end, we propose U-Balance, a supervised approach that leverages behavioral uncertainty to rebalance imbalanced datasets prior to training a safety predictor. U-Balance first trains a GatedMLP-based uncertainty predictor that summarizes each telemetry window into distributional kinematic features and outputs an uncertainty score. It then applies an uncertainty-guided label rebalancing (uLNR) mechanism that probabilistically relabels \textit{safe}-labeled windows with unusually high uncertainty as \textit{unsafe}, thereby enriching the minority class with informative boundary samples without synthesizing new data. Finally, a safety predictor is trained on the rebalanced dataset for safety monitoring. We evaluate U-Balance on a large-scale UAV benchmark with a 46:1 safe-to-unsafe ratio. Results confirm a moderate but significant correlation between behavioral uncertainty and safety. We then identify uLNR as the most effective strategy to exploit uncertainty information, compared to direct early and late fusion. U-Balance achieves a 0.806 F1 score, outperforming the strongest baseline by 14.3 percentage points, while maintaining competitive inference efficiency. Ablation studies confirm that both the GatedMLP-based uncertainty predictor and the uLNR mechanism contribute significantly to U-Balance's effectiveness.

SHAPR (Solo Human-Centred and AI-Assisted Practice) is a framework for research software development that integrates human-centred decision-making with AI-assisted capabilities. While prior work introduced SHAPR as a conceptual framework, this paper focuses on its operationalisation as a structured, traceable, and knowledge-generating approach to AI-assisted research practice. We present a set of interconnected models describing how research activities are organised through iterative cycles (Explore-Build-Use-Evaluate-Learn), how artefacts evolve through development and use, and how empirical evidence is transformed into conceptual knowledge. Central to this process are Structured Knowledge Units (SKUs), which provide modular and reusable representations of insights derived from practice, supporting knowledge accumulation across cycles. The framework introduces evidence and traceability as a cross-cutting mechanism linking human decisions, AI-assisted development, and artefact evolution to enable transparency, reproducibility, and systematic refinement. SHAPR is also positioned as an AI-executable research framework, as its structured processes and documentation can be interpreted by generative AI systems to guide research workflows. Simultaneously, SHAPR supports a continuum of AI involvement, allowing researchers to balance control, learning, and automation across different contexts. Beyond individual workflows, SHAPR is conceptualised as an integrated research system combining LLM workspaces, development environments, cloud storage, and version control to support scalable, knowledge-centred research practices. Overall, SHAPR provides a practical and theoretically grounded foundation for conducting rigorous, transparent, and reproducible research in AI-assisted environments, contributing to the development of scalable and methodologically sound research practices.

This paper presents an experimental platform for studying intentional-state attribution toward a non-humanoid robot. The system combines a simulated robot, realistic task environments, and large language model-based explanatory layers that can express the same behavior in mentalistic, teleological, or mechanistic terms. By holding behavior constant while varying the explanatory frame, the platform provides a controlled way to investigate how language and framing shape the adoption of the intentional stance in robotics.

Accurate parsing of citations is necessary for machine-readable scholarly infrastructure. But, despite sustained interest in this problem, existing evaluation techniques are often not generalizable, based on synthetic data, or not publicly available. We introduce RenoBench, a public domain benchmark for citation parsing, sourced from PDFs released on four publishing ecosystems: SciELO, Redalyc, the Public Knowledge Project, and Open Research Europe. Starting from 161,000 annotated citations, we apply automated validation and feature-based sampling to produce a dataset of 10,000 citations spanning multiple languages, publication types, and platforms. We then evaluate a variety of citation parsing systems and report field-level precision and recall. Our results show strong performance from language models, particularly when fine-tuned. RenoBench enables reproducible, standardized evaluation of citation parsing systems, and provides a foundation for advancing automated citation parsing and metascientific research.

Through an analysis of arXiv papers, we report several shifts in word usage that are likely driven by large language models (LLMs) but have not previously received sufficient attention, such as the increased frequency of "beyond" and "via" in titles and the decreased frequency of "the" and "of" in abstracts. Due to the similarities among different LLMs, experiments show that current classifiers struggle to accurately determine which specific model generated a given text in multi-class classification tasks. Meanwhile, variations across LLMs also result in evolving patterns of word usage in academic papers. By adopting a direct and highly interpretable linear approach and accounting for differences between models and prompts, we quantitatively assess these effects and show that real-world LLM usage is heterogeneous and dynamic.

Air flow modeling at a local scale is essential for applications such as pollutant dispersion modeling or wind farm modeling. To circumvent costly Computational Fluid Dynamics (CFD) computations, deep learning surrogate models have recently emerged as promising alternatives. However, in the context of urban air flow, deep learning models struggle to adapt to the high variations of the urban geometry and to large mesh sizes. To tackle these challenges, we introduce Anchored Branched Steady-state WInd Flow Transformer (AB-SWIFT), a transformer-based model with an internal branched structure uniquely designed for atmospheric flow modeling. We train our model on a specially designed database of atmospheric simulations around randomised urban geometries and with a mixture of unstable, neutral, and stable atmospheric stratifications. Our model reaches the best accuracy on all predicted fields compared to state-of-the-art transformers and graph-based models. Our code and data is available at https://github.com/cerea-daml/abswift.

Large Language Models (LLMs) are increasingly used in math education not only as problem solvers but also as assessors of learners' reasoning. However, it remains unclear whether stronger math problem-solving ability is associated with stronger step-level assessment performance. This study examines that relationship using the GSM8K and MATH subsets of PROCESSBENCH, a human-annotated benchmark for identifying the earliest erroneous step in mathematical reasoning. We evaluate two LLM-based math tutor agent settings, instantiated with GPT-4 and GPT-5, in two independent tasks on the same math problems: solving the original problem and assessing a benchmark-provided solution by predicting the earliest erroneous step. Results show a consistent within-model pattern: assessment accuracy is substantially higher on math problem items the same model solved correctly than on items it solved incorrectly, with statistically significant associations across both models and datasets. At the same time, assessment remains more difficult than direct problem solving, especially on error-present solutions. These findings suggest that math problem-solving expertise supports stronger assessment performance, but reliable step-level diagnosis also requires additional capabilities such as step tracking, monitoring, and precise error localization. The results have implications for the design and evaluation of AI-supported Adaptive Instructional Systems (AISs) for formative assessment in math education.

While language reasoning models excel in many tasks, visual reasoning remains challenging for current large multimodal models (LMMs). As a result, most LMMs default to verbalizing perceptual content into text, a strong limitation for tasks requiring fine-grained spatial and visual understanding. While recent approaches take steps toward thinking with images by invoking tools or generating intermediate images, they either rely on external modules, or incur unnecessary computation by reasoning directly in pixel space. In this paper, we introduce LanteRn, a framework that enables LMMs to interleave language with compact latent visual representations, allowing visual reasoning to occur directly in latent space. LanteRn augments a vision-language transformer with the ability to generate and attend to continuous visual thought embeddings during inference. We train the model in two stages: supervised fine-tuning to ground visual features in latent states, followed by reinforcement learning to align latent reasoning with task-level utility. We evaluate LanteRn on three perception-centric benchmarks (VisCoT, V*, and Blink), observing consistent improvements in visual grounding and fine-grained reasoning. These results suggest that internal latent representations provide a promising direction for more efficient multimodal reasoning.

Explanations are central to improving transparency, trust, and user satisfaction in recommender systems (RS), yet it remains unclear how different explanation formats (visual vs. textual) are suited to users with different personal characteristics (PCs). To this end, we report a within-subject user study (n=54) comparing visual and textual explanations and examine how explanation format and PCs jointly influence perceived control, transparency, trust, and satisfaction in an educational recommender system (ERS). Using robust mixed-effects models, we analyze the moderating effects of a wide range of PCs, including Big Five traits, need for cognition, decision making style, visualization familiarity, and technical expertise. Our results show that a well-designed visual, simple, interactive, selective, easy to understand visualization that clearly and intuitively communicates how user preferences are linked to recommendations, fosters perceived control, transparency, appropriate trust, and satisfaction in the ERS for most users, independent of their PCs. Moreover, we derive a set of guidelines to support the effective design of explanations in ERSs.

We present an efficient algorithm for uniformly sampling from an arbitrary compact body $\mathcal{X} \subset \mathbb{R}^n$ from a warm start under isoperimetry and a natural volume growth condition. Our result provides a substantial common generalization of known results for convex bodies and star-shaped bodies. The complexity of the algorithm is polynomial in the dimension, the Poincaré constant of the uniform distribution on $\mathcal{X}$ and the volume growth constant of the set $\mathcal{X}$.

Large language model (LLM)-based persona agents are rapidly being adopted as scalable proxies for human participants across diverse domains. Yet there is no systematic method for verifying whether a persona agent's responses remain free of contradictions and factual inaccuracies throughout an interaction. A principle from interrogation methodology offers a lens: no matter how elaborate a fabricated identity, systematic interrogation will expose its contradictions. We apply this principle to propose PICon, an evaluation framework that probes persona agents through logically chained multi-turn questioning. PICon evaluates consistency along three core dimensions: internal consistency (freedom from self-contradiction), external consistency (alignment with real-world facts), and retest consistency (stability under repetition). Evaluating seven groups of persona agents alongside 63 real human participants, we find that even systems previously reported as highly consistent fail to meet the human baseline across all three dimensions, revealing contradictions and evasive responses under chained questioning. This work provides both a conceptual foundation and a practical methodology for evaluating persona agents before trusting them as substitutes for human participants. We provide the source code and an interactive demo at: https://kaist-edlab.github.io/picon/

Social learning highlights that learning agents improve not in isolation, but through interaction and structured knowledge exchange with others. When introduced into machine learning, this principle gives rise to social machine learning (SML), where multiple agents collaboratively learn by sharing abstracted knowledge. Federated learning (FL) provides a natural collaboration substrate for this paradigm, yet existing heterogeneous FL approaches often rely on sharing model parameters or intermediate representations, which may expose sensitive information and incur additional overhead. In this work, we propose SoHip (Social Hippocampus Memory Learning), a memory-centric social machine learning framework that enables collaboration among heterogeneous agents via memory sharing rather than model sharing. SoHip abstracts each agent's individual short-term memory from local representations, consolidates it into individual long-term memory through a hippocampus-inspired mechanism, and fuses it with collectively aggregated long-term memory to enhance local prediction. Throughout the process, raw data and local models remain on-device, while only lightweight memory are exchanged. We provide theoretical analysis on convergence and privacy preservation properties. Experiments on two benchmark datasets with seven baselines demonstrate that SoHip consistently outperforms existing methods, achieving up to 8.78% accuracy improvements.

Multimodal Large Language Models (MLLMs) have recently been explored as face verification systems that determine whether two face images are of the same person. Unlike dedicated face recognition systems, MLLMs approach this task through visual prompting and rely on general visual and reasoning abilities. However, the demographic fairness of these models remains largely unexplored. In this paper, we present a benchmarking study that evaluates nine open-source MLLMs from six model families, ranging from 2B to 8B parameters, on the IJB-C and RFW face verification protocols across four ethnicity groups and two gender groups. We measure verification accuracy with the Equal Error Rate and True Match Rate at multiple operating points per demographic group, and we quantify demographic disparity with four FMR-based fairness metrics. Our results show that FaceLLM-8B, the only face-specialised model in our study, substantially outperforms general-purpose MLLMs on both benchmarks. The bias patterns we observe differ from those commonly reported for traditional face recognition, with different groups being most affected depending on the benchmark and the model. We also note that the most accurate models are not necessarily the fairest and that models with poor overall accuracy can appear fair simply because they produce uniformly high error rates across all demographic groups.

The widespread adoption of CT has notably increased the number of detected lung nodules. However, current deep learning methods for classifying benign and malignant nodules often fail to comprehensively integrate global and local features, and most of them have not been validated through clinical trials. To address this, we developed DeepFAN, a transformer-based model trained on over 10K pathology-confirmed nodules and further conducted a multi-reader, multi-case clinical trial to evaluate its efficacy in assisting junior radiologists. DeepFAN achieved diagnostic area under the curve (AUC) of 0.939 (95% CI 0.930-0.948) on an internal test set and 0.954 (95% CI 0.934-0.973) on the clinical trial dataset involving 400 cases across three independent medical institutions. Explainability analysis indicated higher contributions from global than local features. Twelve readers' average performance significantly improved by 10.9% (95% CI 8.3%-13.5%) in AUC, 10.0% (95% CI 8.9%-11.1%) in accuracy, 7.6% (95% CI 6.1%-9.2%) in sensitivity, and 12.6% (95% CI 10.9%-14.3%) in specificity (P<0.001 for all). Nodule-level inter-reader diagnostic consistency improved from fair to moderate (overall k: 0.313 vs. 0.421; P=0.019). In conclusion, DeepFAN effectively assisted junior radiologists and may help homogenize diagnostic quality and reduce unnecessary follow-up of indeterminate pulmonary nodules. Chinese Clinical Trial Registry: ChiCTR2400084624.

Predicting high-dimensional dynamical systems with irregular time steps presents significant challenges for current data-driven algorithms. These irregularities arise from missing data, sparse observations, or adaptive computational techniques, reducing prediction accuracy. To address these limitations, we propose a novel method: a Physics-Spatiotemporal Masked Autoencoder. This method integrates convolutional autoencoders for spatial feature extraction with masked autoencoders optimised for irregular time series, leveraging attention mechanisms to reconstruct the entire physical sequence in a single prediction pass. The model avoids the need for data imputation while preserving physical integrity of the system. Here, 'physics' refers to high-dimensional fields generated by underlying dynamical systems, rather than the enforcement of explicit physical constraints or PDE residuals. We evaluate this approach on multiple simulated datasets and real-world ocean temperature data. The results demonstrate that our method achieves significant improvements in prediction accuracy, robustness to nonlinearities, and computational efficiency over traditional convolutional and recurrent network methods. The model shows potential for capturing complex spatiotemporal patterns without requiring domain-specific knowledge, with applications in climate modelling, fluid dynamics, ocean forecasting, environmental monitoring, and scientific computing.

Repairing faulty quantum circuits is challenging and requires automated solutions. We present QRep, an automated repair approach that iteratively identifies and repairs faults in a circuit. QRep uniformly applies patches across the circuit and assigns each gate a suspiciousness score, reflecting its likelihood of being faulty. It then narrows the search space by prioritising the most suspicious gates in subsequent iterations, increasing the repair efficiency. We evaluated QRep on 40 (real and synthetic) faulty circuits. QRep completely repaired 70% of them, and for the remaining circuits, the actual faulty gate was ranked within the top 44% most suspicious gates, demonstrating the effectiveness of QRep in fault localisation. Compared with two baseline approaches, QRep scales to larger and more complex circuits, up to 13 qubits.

A key capability of modern neural networks is their capacity to simultaneously learn underlying rules and memorize specific facts or exceptions. Yet, theoretical understanding of this dual capability remains limited. We introduce the Rules-and-Facts (RAF) model, a minimal solvable setting that enables precise characterization of this phenomenon by bridging two classical lines of work in the statistical physics of learning: the teacher-student framework for generalization and Gardner-style capacity analysis for memorization. In the RAF model, a fraction $1 - \varepsilon$ of training labels is generated by a structured teacher rule, while a fraction $\varepsilon$ consists of unstructured facts with random labels. We characterize when the learner can simultaneously recover the underlying rule - allowing generalization to new data - and memorize the unstructured examples. Our results quantify how overparameterization enables the simultaneous realization of these two objectives: sufficient excess capacity supports memorization, while regularization and the choice of kernel or nonlinearity control the allocation of capacity between rule learning and memorization. The RAF model provides a theoretical foundation for understanding how modern neural networks can infer structure while storing rare or non-compressible information.

Accurate biodiversity identification from large-scale field data is a foundational problem with direct impact on ecology, conservation, and environmental monitoring. In practice, the core task is taxonomic prediction - inferring order, family, genus, or species from imperfect inputs such as specimen images, DNA barcodes, or both. Existing multimodal methods often treat taxonomy as a flat label space and therefore fail to encode the hierarchical structure of biological classification, which is critical for robustness under noise and missing modalities. We present two end-to-end variants for hierarchy-aware multimodal learning: CLiBD-HiR, which introduces Hierarchical Information Regularization (HiR) to shape embedding geometry across taxonomic levels, yielding structured and noise-robust representations; and CLiBD-HiR-Fuse, which additionally trains a lightweight fusion predictor that supports image-only, DNA-only, or joint inference and is resilient to modality corruption. Across large-scale biodiversity benchmarks, our approach improves taxonomic classification accuracy by over 14 percent compared to strong multimodal baselines, with particularly large gains under partial and corrupted DNA conditions. These results highlight that explicitly encoding biological hierarchy, together with flexible fusion, is key for practical biodiversity foundation models.

The deployment of reconfigurable intelligent surfaces (RISs) introduces new challenges for resource allocation in multi-cell wireless networks, particularly when user loads are uneven across base stations. In this work, we consider RISs as shared infrastructure that must be dynamically assigned among competing base stations, and we address this problem using a simultaneous ascending auction mechanism. To mitigate performance imbalances between cells, we propose a fairness-aware collaborative multi-agent reinforcement learning approach in which base stations adapt their bidding strategies based on both expected utility gains and relative service quality. A centrally computed performance-dependent fairness indicator is incorporated into the agents' observations, enabling implicit coordination without direct inter-base-station communication. Simulation results show that the proposed framework effectively redistributes RIS resources toward weaker-performing cells, substantially improving the rates of the worst-served users while preserving overall throughput. The results demonstrate that fairness-oriented RIS allocation can be achieved through cooperative learning, providing a flexible tool for balancing efficiency and equity in future wireless networks.

With the emergence of AI techniques for depression diagnosis, the conflict between high demand and limited supply for depression screening has been significantly alleviated. Among various modal data, audio-based depression diagnosis has received increasing attention from both academia and industry since audio is the most common carrier of emotion transmission. Unfortunately, audio data also contains User-sensitive Identity Information (ID), which is extremely vulnerable and may be maliciously used during the smart diagnosis process. Among previous methods, the clarification between depression features and sensitive features has always serve as a barrier. It is also critical to the problem for introducing a safe encryption methodology that only encrypts the sensitive features and a powerful classifier that can correctly diagnose the depression. To track these challenges, by leveraging adversarial loss-based Subspace Decomposition, we propose a first practical framework \name presented for Trustable Audio Affective Computing, to perform automated depression detection through audio within a trustable environment. The key enablers of TAAC are Differentiating Features Subspace Decompositor (DFSD), Flexible Noise Encryptor (FNE) and Staged Training Paradigm, used for decomposition, ID encryption and performance enhancement, respectively. Extensive experiments with existing encryption methods demonstrate our framework's preeminent performance in depression detection, ID reservation and audio reconstruction. Meanwhile, the experiments across various setting demonstrates our model's stability under different encryption strengths. Thus proving our framework's excellence in Confidentiality, Accuracy, Traceability, and Adjustability.

Translating natural language to SQL for data retrieval has become more accessible thanks to code generation LLMs. But how hard is it to generate SQL code? While databases can become unbounded in complexity, the complexity of queries is bounded by real life utility and human needs. With a sample of 376 databases, we show that SQL queries, as translations of natural language questions are finite in practical complexity. There is no clear monotonic relationship between increases in database table count and increases in complexity of SQL queries. In their template forms, SQL queries follow a Power Law-like distribution of frequency where 70% of our tested queries can be covered with just 13% of all template types, indicating that the high majority of SQL queries are predictable. This suggests that while LLMs for code generation can be useful, in the domain of database access, they may be operating in a narrow, highly formulaic space where templates could be safer, cheaper, and auditable.

On-policy distillation (OPD) is appealing for large language model (LLM) post-training because it evaluates teacher feedback on student-generated rollouts rather than fixed teacher traces. In long-horizon settings, however, the common sampled-token variant is fragile: it reduces distribution matching to a one-token signal and becomes increasingly unreliable as rollouts drift away from prefixes the teacher commonly visits. We revisit OPD from the estimator and implementation sides. Theoretically, token-level OPD is biased relative to sequence-level reverse-KL, but it has a much tighter worst-case variance bound; our toy study shows the same tradeoff empirically, with stronger future-reward coupling producing higher gradient variance and less stable learning. Empirically, we identify three failure modes of sampled-token OPD: an imbalanced one-token signal, unreliable teacher guidance on student-generated prefixes, and distortions caused by tokenizer or special-token mismatch. We address these issues with teacher top-K local support matching, implemented as truncated reverse-KL with top-p rollout sampling and special-token masking. Across single-task math reasoning and multi-task agentic-plus-math training, this objective yields more stable optimization and better downstream performance than sampled-token OPD.

Saccharomyces cerevisiae is a cornerstone organism in industrial biotechnology, valued for its genetic tractability and robust fermentative capacity. Accurately predicting biomass flux across diverse environmental and genetic perturbations remains a significant challenge for rational strain design. We present a computational framework combining the Yeast9 genome-scale metabolic model with machine learning and optimization to predict, interpret, and enhance biomass flux. Flux balance analysis generated 2,000 flux profiles by varying glucose, oxygen, and ammonium uptake rates. Random Forest and XGBoost regressors achieved R2 of 0.99989 and 0.9990, respectively. A variational autoencoder revealed four distinct metabolic clusters, and SHAP analysis identified glycolysis, the TCA cycle, and lipid biosynthesis as key biomass determinants. In silico overexpression achieved a biomass flux of 0.979 gDW/hr, while Bayesian optimization of nutrient constraints produced a 12-fold increase (0.0858 to 1.041 gDW/hr). A generative adversarial network proposed stoichiometrically feasible novel flux configurations. This framework demonstrates how genome-scale simulation, interpretable ML, and generative modeling can advance yeast metabolic engineering.

We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch with a hybrid VQ-FSQ quantization scheme. In human evaluations conducted by native speakers, Voxtral TTS is preferred for multilingual voice cloning due to its naturalness and expressivity, achieving a 68.4\% win rate over ElevenLabs Flash v2.5. We release the model weights under a CC BY-NC license.

Automated incident management is critical for microservice reliability. While recent unified frameworks leverage multimodal data for joint optimization, they unrealistically assume perfect data completeness. In practice, network fluctuations and agent failures frequently cause missing modalities. Existing approaches relying on static placeholders introduce imputation noise that masks anomalies and degrades performance. To address this, we propose ARMOR, a robust self-supervised framework designed for missing modality scenarios. ARMOR features: (i) a modality-specific asymmetric encoder that isolates distribution disparities among metrics, logs, and traces; and (ii) a missing-aware gated fusion mechanism utilizing learnable placeholders and dynamic bias compensation to prevent cross-modal interference from incomplete inputs. By employing self-supervised auto-regression with mask-guided reconstruction, ARMOR jointly optimizes anomaly detection (AD), failure triage (FT), and root cause localization (RCL). AD and RCL require no fault labels, while FT relies solely on failure-type annotations for the downstream classifier. Extensive experiments demonstrate that ARMOR achieves state-of-the-art performance under complete data conditions and maintains robust diagnostic accuracy even with severe modality loss.

We study narrative coherence in visually grounded stories by comparing human-written narratives with those generated by vision-language models (VLMs) on the Visual Writing Prompts corpus. Using a set of metrics that capture different aspects of narrative coherence, including coreference, discourse relation types, topic continuity, character persistence, and multimodal character grounding, we compute a narrative coherence score. We find that VLMs show broadly similar coherence profiles that differ systematically from those of humans. In addition, differences for individual measures are often subtle, but they become clearer when considered jointly. Overall, our results indicate that, despite human-like surface fluency, model narratives exhibit systematic differences from those of humans in how they organise discourse across a visually grounded story. Our code is available at https://github.com/GU-CLASP/coherence-driven-humans.

To date, there is little research on how to design back marks to best support individual-level monitoring of uniform looking species like pigs. With the recent surge of machine learning-based monitoring solutions, there is a particular need for guidelines on the design of marks that can be effectively recognised by such algorithms. This study provides valuable insights on effective back mark design, based on the analysis of a machine learning model, trained to distinguish pigs via their back marks. Specifically, a neural network of type ResNet-50 was trained to classify ten pigs with unique back marks. The analysis of the model's predictions highlights the significance of certain design choices, even in controlled settings. Most importantly, the set of back marks must be designed such that each mark remains unambiguous under conditions of motion blur, diverse view angles and occlusions, caused by animal behaviour. Further, the back mark design must consider data augmentation strategies commonly employed during model training, like colour, flip and crop augmentations. The generated insights can support individual-level monitoring in future studies and real-world applications by optimizing back mark design.

Many Cyber Physical System (CPS) work in a safety-critical environment, where correct execution, reliability and trustworthiness are essential. Signal Temporal Logic (STL) provides a formal framework for checking safety-critical CPS. However, static verification of STL is undecidable in general, except when we want to verify using run-time-based methods, which have limitations. We propose Synchronous Signal Temporal Logic (SSTL), a decidable fragment of STL, which admits static safety and liveness property verification. In SSTL, we assume that a signal is sampled at fixed discrete steps, called ticks, and then propose a hypothesis, called the Signal Invariance Hypothesis (SIH), which is inspired by a similar hypothesis for synchronous programs. We define the syntax and semantics of SSTL and show that SIH is a necessary and sufficient condition for equivalence between an STL formula and its SSTL counterpart. By translating SSTL to LTL_P (LTL defined over predicates), we enable decidable model checking using the SPIN model checker. We demonstrate the approach on a 33-node human heart model and other case studies.

Long-term behavioral monitoring of individual animals is crucial for studying behavioral changes that occur over different time scales, especially for conservation and evolutionary biology. Computer vision methods have proven to benefit biodiversity monitoring, but automated behavior monitoring in wild populations remains challenging. This stems from the lack of datasets that cover a range of computer vision tasks necessary to extract biologically meaningful measurements of individual animals. Here, we introduce such a dataset (CHIRP) with a new method (CORVID) for individual re-identification of wild birds. The CHIRP (Combining beHaviour, Individual Re-identification and Postures) dataset is curated from a long-term population of wild Siberian jays studied in Swedish Lapland, supporting re-identification (re-id), action recognition, 2D keypoint estimation, object detection, and instance segmentation. In addition to traditional task-specific benchmarking, we introduce application-specific benchmarking with biologically relevant metrics (feeding rates, co-occurrence rates) to evaluate the performance of models in real-world use cases. Finally, we present CORVID (COlouR-based Video re-ID), a novel pipeline for individual identification of birds based on the segmentation and classification of colored leg rings, a widespread approach for visual identification of individual birds. CORVID offers a probability-based id tracking method by matching the detected combination of color rings with a database. We use application-specific benchmarking to show that CORVID outperforms state-of-the-art re-id methods. We hope this work offers the community a blueprint for curating real-world datasets from ethically approved biological studies to bridge the gap between computer vision research and biological applications.

Neuroevolution automates the complex task of neural network design but often ignores the inherent adversarial fragility of evolved models which is a barrier to adoption in safety-critical scenarios. While robust training methods have received significant attention, the design of architectures exhibiting intrinsic robustness remains largely unexplored. In this paper, we propose NERO-Net, a neuroevolutionary approach to design convolutional neural networks better equipped to resist adversarial attacks. Our search strategy isolates architectural influence on robustness by avoiding adversarial training during the evolutionary loop. As such, our fitness function promotes candidates that, even trained with standard (non-robust) methods, achieve high post-attack accuracy without sacrificing the accuracy on clean samples. We assess NERO-Net on CIFAR-10 with a specific focus on $L_\infty$-robustness. In particular, the fittest individual emerged from evolutionary search with 33% accuracy against FGSM, used as an efficient estimator for robustness during the search phase, while maintaining 87% clean accuracy. Further standard training of this individual boosted these metrics to 47% adversarial and 93% clean accuracy, suggesting inherent architectural robustness. Adversarial training brings the overall accuracy of the model up to 40% against AutoAttack.

The use of hyperspectral imaging (HSI) in autonomous driving (AD), while promising, faces many challenges related to the specifics and requirements of this application domain. On the one hand, non-controlled and variable lighting conditions, the wide depth-of-field ranges, and dynamic scenes with fast-moving objects. On the other hand, the requirements for real-time operation and the limited computational resources of embedded platforms. The combination of these factors determines both the criteria for selecting appropriate HSI technologies and the development of custom vision algorithms that leverage the spectral and spatial information obtained from the sensors. In this article, we analyse several techniques explored in the research of HSI-based vision systems with application to AD, using as an example results obtained from experiments using data from the most recent version of the HSI-Drive dataset.

We propose a method for constructing distribution-free prediction intervals in nonparametric instrumental variable regression (NPIV), with finite-sample coverage guarantees. Building on the conditional guarantee framework in conformal inference, we reformulate conditional coverage as marginal coverage over a class of IV shifts $\mathcal{F}$. Our method can be combined with any NPIV estimator, including sieve 2SLS and other machine-learning-based NPIV methods such as neural networks minimax approaches. Our theoretical analysis establishes distribution-free, finite-sample coverage over a practitioner-chosen class of IV shifts.

Accurate Network Traffic Classification (NTC) is increasingly constrained by limited labeled data and strict privacy requirements. While Network Traffic Generation (NTG) provides an effective means to mitigate data scarcity, conventional generative methods struggle to model the complex temporal dynamics of modern traffic or/and often incur significant computational cost. In this article, we address the NTG task using lightweight Generative Artificial Intelligence (GenAI) architectures, including transformer-based, state-space, and diffusion models designed for practical deployment. We conduct a systematic evaluation along four axes: (i) (synthetic) traffic fidelity, (ii) synthetic-only training, (iii) data augmentation under low-data regimes, and (iv) computational efficiency. Experiments on two heterogeneous datasets show that lightweight GenAI models preserve both static and temporal traffic characteristics, with transformer and state-space models closely matching real distributions across a complete set of fidelity metrics. Classifiers trained solely on synthetic traffic achieve up to 87% F1-score on real data. In low-data settings, GenAI-driven augmentation improves NTC performance by up to +40%, substantially reducing the gap with full-data training. Overall, transformer-based models provide the best trade-off between fidelity and efficiency, enabling high-quality, privacy-aware traffic synthesis with modest computational overhead.

In recent years, fake news detection has received increasing attention in public debate and scientific research. Despite advances in detection techniques, the production and spread of false information have become more sophisticated, driven by Large Language Models (LLMs) and the amplification power of social media. We present a critical assessment of 12 representative fake news detection approaches, spanning traditional machine learning, deep learning, transformers, and specialized cross-domain architectures. We evaluate these methods on 10 publicly available datasets differing in genre, source, topic, and labeling rationale. We address text-only English fake news detection as a binary classification task by harmonizing labels into "Real" and "Fake" to ensure a consistent evaluation protocol. We acknowledge that label semantics vary across datasets and that harmonization inevitably removes such semantic nuances. Each dataset is treated as a distinct domain. We conduct in-domain, multi-domain and cross-domain experiments to simulate real-world scenarios involving domain shift and out-of-distribution data. Fine-tuned models perform well in-domain but struggle to generalize. Cross-domain architectures can reduce this gap but are data-hungry, while LLMs offer a promising alternative through zero- and few-shot learning. Given inherent dataset confounds and possible pre-training exposure, results should be interpreted as robustness evaluations within this English, text-only protocol.

Object detectors deployed in safety-critical environments can fail silently, e.g. missing pedestrians, workers, or other safety-critical objects without emitting any warning. Traditional Out Of Distribution (OOD) detection methods focus on identifying unfamiliar inputs, but do not directly predict functional failures of the detector itself. We introduce Knowledge Guided Failure Prediction (KGFP), a representation-based monitoring framework that treats missed safety-critical detections as anomalies to be detected at runtime. KGFP measures semantic misalignment between internal object detector features and visual foundation model embeddings using a dual-encoder architecture with an angular distance metric. A key property is that when either the detector is operating outside its competence or the visual foundation model itself encounters novel inputs, the two embeddings diverge, producing a high-angle signal that reliably flags unsafe images. We compare our novel KGFS method to baseline OOD detection methods. On COCO person detection, applying KGFP as a selective-prediction gate raises person recall among accepted images from 64.3% to 84.5% at 5% False Positive Rate (FPR), and maintains strong performance across six COCO-O visual domains, outperforming OOD baselines by large margins. Our code, models, and features are published at https://gitlab.cc-asp.fraunhofer.de/iosb_public/KGFP.

As the Web transitions from static retrieval to generative interaction, the escalating environmental footprint of Large Language Models (LLMs) presents a critical sustainability challenge. Current paradigms indiscriminately apply computation-intensive strategies like Chain-of-Thought (CoT) to billions of daily queries, causing LLM overthinking, a redundancy that amplifies carbon emissions and operational barriers. This inefficiency directly undermines UN Sustainable Development Goals 13 (Climate Action) and 10 (Reduced Inequalities) by hindering equitable AI access in resource-constrained regions. To address this, we introduce EcoThink, an energy-aware adaptive inference framework designed to reconcile high-performance AI intelligence with environmental responsibility. EcoThink employs a lightweight, distillation-based router to dynamically assess query complexity, skipping unnecessary reasoning for factoid retrieval while reserving deep computation for complex logic. Extensive evaluations across 9 diverse benchmarks demonstrate that EcoThink reduces inference energy by 40.4% on average (up to 81.9% for web knowledge retrieval) without statistically significant performance loss. By mitigating algorithmic waste, EcoThink offers a scalable path toward a sustainable, inclusive, and energy-efficient generative AI Agent.

Accurate short-term air-quality forecasting is essential for public health protection and urban management, yet many recent forecasting frameworks rely on complex, data-intensive, and computationally demanding models. This study investigates whether lightweight and interpretable forecasting approaches can provide competitive performance for hourly PM2.5 prediction in Beijing, China. Using multi-year pollutant and meteorological time-series data, we developed a leakage-aware forecasting workflow that combined chronological data partitioning, preprocessing, feature selection, and exogenous-driver modeling under the Perfect Prognosis setting. Three forecasting families were evaluated: SARIMAX, Facebook Prophet, and NeuralProphet. To assess practical deployment behavior, the models were tested under two adaptive regimes: weekly walk-forward refitting and frozen forecasting with online residual correction. Results showed clear differences in both predictive accuracy and computational efficiency. Under walk-forward refitting, Facebook Prophet achieved the strongest completed performance, with an MAE of $37.61$ and an RMSE of $50.10$, while also requiring substantially less execution time than NeuralProphet. In the frozen-model regime, online residual correction improved Facebook Prophet and SARIMAX, with corrected SARIMAX yielding the lowest overall error (MAE $32.50$; RMSE $46.85$). NeuralProphet remained less accurate and less stable across both regimes, and residual correction did not improve its forecasts. Notably, corrected Facebook Prophet reached nearly the same error as its walk-forward counterpart while reducing runtime from $15$ min $21.91$ sec to $46.60$ sec. These findings show that lightweight additive forecasting strategies can remain highly competitive for urban air-quality prediction, offering a practical balance between accuracy, interpretability, ...

Recent strategies for low-resource machine translation rely on LLMs to generate synthetic data from higher-resource languages. We find that this method fails for Romansh, because LLMs tend to confuse its 6 distinct language varieties. Our experiments show that instead, the direction of data augmentation should be aligned with the resource gradient between source and target language. This approach surpasses Gemini 3 Pro in the lowest-resource variety of Romansh by 23 BLEU. A human evaluation confirms that our experiments yield the first model that generates fluent translations in the individual Romansh varieties.

Model retraining is usually treated as an ongoing maintenance task. But as Harrison Katz now argues, retraining can be better understood as approximate Bayesian inference under computational constraints. The gap between a continuously updated belief state and your frozen deployed model is "learning debt," and the retraining decision is a cost minimization problem with a threshold that falls out of your loss function. In this article Katz provides a decision-theoretic framework for retraining policies. The result is evidence-based triggers that replace calendar schedules and make governance auditable. For readers less familiar with the Bayesian and decision-theoretic language, key terms are defined in a glossary at the end of the article.

Transfer learning is a crucial concept within deep learning that allows artificial neural networks to benefit from a large pre-training data basis when confronted with a task of limited data. Despite its ubiquitous use and clear benefits, there are still many open questions regarding the inner workings of transfer learning and, in particular, regarding the understanding of when and how well it works. To that extent, we perform a rigorous study focusing on audio-to-audio transfer learning, in which we pre-train various model states on (ontology-based) subsets of AudioSet and fine-tune them on three computer audition tasks, namely acoustic scene recognition, bird activity recognition, and speech command recognition. We report that increasing the number of samples and classes in the pre-training data both have a positive impact on transfer learning. This is, however, generally surpassed by similarity between pre-training and the downstream task, which can lead the model to learn comparable features.

Understanding directed temporal interactions in multivariate time series is essential for interpreting complex dynamical systems and the predictive models trained on them. We present Causal-INSIGHT, a model-agnostic, post-hoc interpretation framework for extracting model-implied (predictor-dependent), directed, time-lagged influence structure from trained temporal predictors. Rather than inferring causal structure at the level of the data-generating process, Causal-INSIGHT analyzes how a fixed, pre-trained predictor responds to systematic, intervention-inspired input clamping applied at inference time. From these responses, we construct directed temporal influence signals that reflect the dependencies the predictor relies on for prediction, and introduce Qbic, a sparsity-aware graph selection criterion that balances predictive fidelity and structural complexity without requiring ground-truth graph labels. Experiments across synthetic, simulated, and realistic benchmarks show that Causal-INSIGHT generalizes across diverse backbone architectures, maintains competitive structural accuracy, and yields significant improvements in temporal delay localization when applied to existing predictors.

A growing body of literature has focused on predicting wildfire occurrence using machine learning methods, capitalizing on high-resolution data and fire predictors that canonical process-based frameworks largely ignore. Standard evaluation metrics for an ML classifier, while important, provide a potentially limited measure of the model's operational performance for the Fire Danger Index (FDI) forecast. Furthermore, model evaluation is frequently conducted without adequately accounting for false positive rates, despite their critical relevance in operational contexts. In this paper, we revisit the daily FDI model evaluation paradigm and propose a novel method for evaluating a forest fire forecasting model that is aligned with real-world decision-making. Furthermore, we systematically assess performance in accurately predicting fire activity and the false positives (false alarms). We further demonstrate that an ensemble of ML models improves both fire identification and reduces false positives.

We study statistical estimation in a student--teacher setting, where predictions from a pre-trained teacher are used to guide a student model. A standard approach is to train the student to directly match the teacher's outputs, which we refer to as student soft matching (SM). This approach directly propagates any systematic bias or mis-specification present in the teacher, thereby degrading the student's predictions. We propose and analyze an alternative scheme, known as residual-as-teacher (RaT), in which the teacher is used to estimate residuals in the student's predictions. Our analysis shows how the student can thereby emulate a proximal gradient scheme for solving an oracle optimization problem, and this provably reduces the effect of teacher bias. For general student--teacher pairs, we establish non-asymptotic excess risk bounds for any RaT fixed point, along with convergence guarantees for the student-teacher iterative scheme. For kernel-based student--teacher pairs, we prove a sharp separation: the RaT method achieves the minimax-optimal rate, while the SM method incurs constant prediction error for any sample size. Experiments on both synthetic data and ImageNette classification under covariate shift corroborate our theoretical findings.

Zero-shot reinforcement learning (RL) algorithms aim to learn a family of policies from a reward-free dataset, and recover optimal policies for any reward function directly at test time. Naturally, the quality of the pretraining dataset determines the performance of the recovered policies across tasks. However, pre-collecting a relevant, diverse dataset without prior knowledge of the downstream tasks of interest remains a challenge. In this work, we study $\textit{online}$ zero-shot RL for quadrupedal control on real robotic systems, building upon the Forward-Backward (FB) algorithm. We observe that undirected exploration yields low-diversity data, leading to poor downstream performance and rendering policies impractical for direct hardware deployment. Therefore, we introduce FB-MEBE, an online zero-shot RL algorithm that combines an unsupervised behavior exploration strategy with a regularization critic. FB-MEBE promotes exploration by maximizing the entropy of the achieved behavior distribution. Additionally, a regularization critic shapes the recovered policies toward more natural and physically plausible behaviors. We empirically demonstrate that FB-MEBE achieves and improved performance compared to other exploration strategies in a range of simulated downstream tasks, and that it renders natural policies that can be seamlessly deployed to hardware without further finetuning. Videos and code available on our website.

Motion planning in dynamic urban environments requires balancing immediate safety with long-term goals. While diffusion models effectively capture multi-modal decision-making, existing approaches treat trajectories as monolithic entities, overlooking heterogeneous temporal dependencies where near-term plans are constrained by instantaneous dynamics and far-term plans by navigational goals. To address this, we propose Temporally Decoupled Diffusion Model (TDDM), which reformulates trajectory generation via a noise-as-mask paradigm. By partitioning trajectories into segments with independent noise levels, we implicitly treat high noise as information voids and weak noise as contextual cues. This compels the model to reconstruct corrupted near-term states by leveraging internal correlations with better-preserved temporal contexts. Architecturally, we introduce a Temporally Decoupled Adaptive Layer Normalization (TD-AdaLN) to inject segment-specific timesteps. During inference, our Asymmetric Temporal Classifier-Free Guidance utilizes weakly noised far-term priors to guide immediate path generation. Evaluations on the nuPlan benchmark show TDDM approaches or exceeds state-of-the-art baselines, particularly excelling in the challenging Test14-hard subset.

Detecting when a language model is wrong without ground truth labels is a fundamental challenge for safe deployment. Existing approaches rely on a model's own uncertainty -- such as token entropy or confidence scores -- but these signals fail critically on the most dangerous failure mode: confident errors, where a model is wrong but certain. In this work we introduce cross-model disagreement as a correctness indicator -- a simple, training-free signal that can be dropped into existing production systems, routing pipelines, and deployment monitoring infrastructure without modification. Given a model's generated answer, cross-model disagreement computes how surprised or uncertain a second verifier model is when reading that answer via a single forward pass. No generation from the verifying model is required, and no correctness labels are needed. We instantiate this principle as Cross-Model Perplexity (CMP), which measures the verifying model's surprise at the generating model's answer tokens, and Cross-Model Entropy (CME), which measures the verifying model's uncertainty at those positions. Both CMP and CME outperform within-model uncertainty baselines across benchmarks spanning reasoning, retrieval, and mathematical problem solving (MMLU, TriviaQA, and GSM8K). On MMLU, CMP achieves a mean AUROC of 0.75 against a within-model entropy baseline of 0.59. These results establish cross-model disagreement as a practical, training-free approach to label-free correctness estimation, with direct applications in deployment monitoring, model routing, selective prediction, data filtering, and scalable oversight of production language model systems.

We introduce and solve a teacher-student formulation of the symmetric binary Perceptron, turning a traditionally storage-oriented model into a planted inference problem with a guaranteed solution at any sample density. We adapt the formulation of the symmetric Perceptron which traditionally considers either the u-shaped potential or the rectangular one, by including labels in both regions. With this formulation, we analyze both the Bayes-optimal regime at for noise-less examples and the effect of thermal noise under two different potential/classification rules. Using annealed and quenched free-entropy calculations in the high-dimensional limit, we map the phase diagram in the three control parameters, namely the sample density $α$, the distance between the origin and one of the symmetric hyperplanes $κ$ and temperature $T$, and identify a robust scenario where learning is organized by a second-order instability that creates teacher-correlated suboptimal states, followed by a first-order transition to full alignment. We show how this structure depends on the choice of potential, the interplay between metastability of the suboptimal solution and its melting towards the planted configuration, which is relevant for Monte Carlo-based optimization algorithms.

The rise of micro-videos has reshaped how misinformation spreads, amplifying its speed, reach, and impact on public trust. Existing benchmarks typically focus on a single deception type, overlooking the diversity of real-world cases that involve multimodal manipulation, AI-generated content, cognitive bias, and out-of-context reuse. Meanwhile, most detection models lack fine-grained attribution, limiting interpretability and practical utility. To address these gaps, we introduce WildFakeBench, a large-scale benchmark of over 10,000 real-world micro-videos covering diverse misinformation types and sources, each annotated with expert-defined attribution labels. Building on this foundation, we develop FakeAgent, a Delphi-inspired multi-agent reasoning framework that integrates multimodal understanding with external evidence for attribution-grounded analysis. FakeAgent jointly analyzes content and retrieved evidence to identify manipulation, recognize cognitive and AI-generated patterns, and detect out-of-context misinformation. Extensive experiments show that FakeAgent consistently outperforms existing MLLMs across all misinformation types, while WildFakeBench provides a realistic and challenging testbed for advancing explainable micro-video misinformation detection. Data and code are available at: https://github.com/Aiyistan/FakeAgent.

Recent developments in text classification using Large Language Models (LLMs) in the social sciences suggest that costs can be cut significantly, while performance can sometimes rival existing computational methods. However, with a wide variance in performance in current tests, we move to the question of how to maximize performance. In this paper, we focus on prompt context as a possible avenue for increasing accuracy by systematically varying three aspects of prompt engineering: label descriptions, instructional nudges, and few shot examples. Across two different examples, our tests illustrate that a minimal increase in prompt context yields the highest increase in performance, while further increases in context only tend to yield marginal performance increases thereafter. Alarmingly, increasing prompt context sometimes decreases accuracy. Furthermore, our tests suggest substantial heterogeneity across models, tasks, and batch size, underlining the need for individual validation of each LLM coding task rather than reliance on general rules.

Large Language Models (LLMs) have demonstrated remarkable proficiency in English mathematical reasoning, yet a significant performance disparity persists in multilingual contexts, largely attributed to deficiencies in language understanding. To bridge this gap, we introduce Translation-Augmented Policy Optimization (TAPO), a novel reinforcement learning framework built upon GRPO. TAPO enforces an explicit alignment strategy where the model leverages English as a pivot and follows an understand-then-reason paradigm. Crucially, we employ a step-level relative advantage mechanism that decouples understanding from reasoning, allowing the integration of translation quality rewards without introducing optimization conflicts. Extensive experiments reveal that TAPO effectively synergizes language understanding with reasoning capabilities and is compatible with various models. It outperforms baseline methods in both multilingual mathematical reasoning and translation tasks, while generalizing well to unseen languages and out-of-domain tasks.

Semantic world models enable embodied agents to reason about objects, relations, and spatial context beyond purely geometric representations. In Organic Computing, such models are a key enabler for objective-driven self-adaptation under uncertainty and resource constraints. The core challenge is to acquire observations maximising model quality and downstream usefulness within a limited action budget. Semantic scene graphs (SSGs) provide a structured and compact representation for this purpose. However, constructing them within a finite action horizon requires exploration strategies that trade off information gain against navigation cost and decide when additional actions yield diminishing returns. This work presents a modular navigation component for Embodied Semantic Scene Graph Generation and modernises its decision-making by replacing the policy-optimisation method and revisiting the discrete action formulation. We study compact and finer-grained, larger discrete motion sets and compare a single-head policy over atomic actions with a factorised multi-head policy over action components. We evaluate curriculum learning and optional depth-based collision supervision, and assess SSG completeness, execution safety, and navigation behaviour. Results show that replacing the optimisation algorithm alone improves SSG completeness by 21\% relative to the baseline under identical reward shaping. Depth mainly affects execution safety (collision-free motion), while completeness remains largely unchanged. Combining modern optimisation with a finer-grained, factorised action representation yields the strongest overall completeness--efficiency trade-off.

A prevailing assumption in machine learning is that model correctness must be enforced after the fact. We observe that the properties determining whether an AI model is numerically stable, computationally correct, or consistent with a physical domain do not necessarily demand post hoc enforcement. They can be verified at design time, before training begins, at marginal computational cost, with particular relevance to models deployed in high-leverage decision support and scientifically constrained settings. These properties share a specific algebraic structure: they are expressible as constraints over finitely generated abelian groups $\mathbb{Z}^n$, where inference is decidable in polynomial time and the principal type is unique. A framework built on this observation composes three prior results (arXiv:2603.16437, arXiv:2603.17627, arXiv:2603.18104): a dimensional type system carrying arbitrary annotations as persistent codata through model elaboration; a program hypergraph that infers Clifford algebra grade and derives geometric product sparsity from type signatures alone; and an adaptive domain model architecture preserving both invariants through training via forward-mode coeffect analysis and exact posit accumulation. We believe this composition yields a novel information-theoretic result: Hindley-Milner unification over abelian groups computes the maximum a posteriori hypothesis under a computable restriction of Solomonoff's universal prior, placing the framework's type inference on the same formal ground as universal induction. We compare four contemporary approaches to AI reliability and show that each imposes overhead that can compound across deployments, layers, and inference requests. This framework eliminates that overhead by construction.

Large language models (LLMs) increasingly rely on explicit chain-of-thought (CoT) reasoning to solve complex tasks, yet the safety of the reasoning process itself remains largely unaddressed. Existing work on LLM safety focuses on content safety--detecting harmful, biased, or factually incorrect outputs -- and treats the reasoning chain as an opaque intermediate artifact. We identify reasoning safety as an orthogonal and equally critical security dimension: the requirement that a model's reasoning trajectory be logically consistent, computationally efficient, and resistant to adversarial manipulation. We make three contributions. First, we formally define reasoning safety and introduce a nine-category taxonomy of unsafe reasoning behaviors, covering input parsing errors, reasoning execution errors, and process management errors. Second, we conduct a large-scale prevalence study annotating 4111 reasoning chains from both natural reasoning benchmarks and four adversarial attack methods (reasoning hijacking and denial-of-service), confirming that all nine error types occur in practice and that each attack induces a mechanistically interpretable signature. Third, we propose a Reasoning Safety Monitor: an external LLM-based component that runs in parallel with the target model, inspects each reasoning step in real time via a taxonomy-embedded prompt, and dispatches an interrupt signal upon detecting unsafe behavior. Evaluation on a 450-chain static benchmark shows that our monitor achieves up to 84.88\% step-level localization accuracy and 85.37\% error-type classification accuracy, outperforming hallucination detectors and process reward model baselines by substantial margins. These results demonstrate that reasoning-level monitoring is both necessary and practically achievable, and establish reasoning safety as a foundational concern for the secure deployment of large reasoning models.

Long-horizon tabletop games pose a distinct systems challenge for robotics: small perceptual or execution errors can invalidate accumulated task state, propagate across decision-making modules, and ultimately derail interaction. This paper studies how to maintain internal state consistency in turn-based, multi-human robotic tabletop games through deliberate system design rather than isolated component improvement. Using Mahjong as a representative long-horizon setting, we present an integrated architecture that explicitly maintains perceptual, execution, and interaction state, partitions high-level semantic reasoning from time-critical perception and control, and incorporates verified action primitives with tactile-triggered recovery to prevent premature state corruption. We further introduce interaction-level monitoring mechanisms to detect turn violations and hidden-information breaches that threaten execution assumptions. Beyond demonstrating complete-game operation, we provide an empirical characterization of failure modes, recovery effectiveness, cross-module error propagation, and hardware-algorithm trade-offs observed during deployment. Our results show that explicit partitioning, monitored state transitions, and recovery mechanisms are critical for sustaining executable consistency over extended play, whereas monolithic or unverified pipelines lead to measurable degradation in end-to-end reliability. The proposed system serves as an empirical platform for studying system-level design principles in long-horizon, turn-based interaction.

On-device Vision-Language Models (VLMs) promise data privacy via local execution. However, we show that the architectural shift toward Dynamic High-Resolution preprocessing (e.g., AnyRes) introduces an inherent algorithmic side-channel. Unlike static models, dynamic preprocessing decomposes images into a variable number of patches based on their aspect ratio, creating workload-dependent inputs. We demonstrate a dual-layer attack framework against local VLMs. In Tier 1, an unprivileged attacker can exploit significant execution-time variations using standard unprivileged OS metrics to reliably fingerprint the input's geometry. In Tier 2, by profiling Last-Level Cache (LLC) contention, the attacker can resolve semantic ambiguity within identical geometries, distinguishing between visually dense (e.g., medical X-rays) and sparse (e.g., text documents) content. By evaluating state-of-the-art models such as LLaVA-NeXT and Qwen2-VL, we show that combining these signals enables reliable inference of privacy-sensitive contexts. Finally, we analyze the security engineering trade-offs of mitigating this vulnerability, reveal substantial performance overhead with constant-work padding, and propose practical design recommendations for secure Edge AI deployments.

In this applied paper, we address the difficult open problem of when to discharge patients from the Intensive Care Unit. This can be conceived as an optimal stopping scenario with three added challenges: 1) the evaluation of a stopping strategy from observational data is itself a complex causal inference problem, 2) the composite objective is to minimize the length of intervention and maximize the outcome, but the two cannot be collapsed to a single dimension, and 3) the recording of variables stops when the intervention is discontinued. Our contributions are two-fold. First, we generalize the implementation of the g-formula Python package, providing a framework to evaluate stopping strategies for problems with the aforementioned structure, including positivity and coverage checks. Second, with a fully open-source pipeline, we apply this approach to MIMIC-IV, a public ICU dataset, demonstrating the potential for strategies that improve upon current care.

Every software organization faces the onboarding challenge: helping newcomers navigate complex codebases, compensate for insufficient documentation, and comprehend code they did not author. Expert walkthroughs are among the most effective forms of support, yet they are expensive, repetitive, and do not scale. We present Lacy, a hybrid human-AI onboarding system that captures expert mentoring in reusable code tours-to our knowledge, the first hybrid approach combining AI-generated content with expert curation in code tours. Our design is grounded in requirements derived from 20+ meetings, surveys, and interviews across a year-long industry partnership with Beko. Supporting features include Voice-to-Tour capture, comprehension quizzes, podcasts, and a dashboard. We deployed Lacy on Beko's production environment and conducted a controlled study on a legacy finance system (30K+ LOC). Learners using expert-guided tours achieved 83% quiz scores versus 57% for AI-only tours, preferred tours over traditional self-study, and reported they would need fewer expert consultations. Experts found tour creation less burdensome than live walkthroughs. Beko has since adopted Lacy for organizational onboarding, and we release our code and study instruments as a replication package.

Quantization techniques such as BitsAndBytes, AWQ, and GPTQ are widely used as a standard method in deploying large language models but often degrades accuracy when using low-bit representations, e.g., 4 bits. Low-rank correction methods (e.g., LQER, QERA, ASER) has been proposed to mitigate this issue, however, they restore all layers and insert error-correction modules into every decoder block, which increases latency and memory overhead. To address this limitation, we propose GlowQ, a group-shared low-rank approximation for quantized LLMs that caches a single shared right factor per input-sharing group and restores only the groups or layers that yield the highest accuracy benefit. GlowQ computes the high-precision projection once per input-sharing group and reuses it across its modules, reducing parameter and memory overhead, and retaining the expressivity of layer-specific corrections. We also propose a selective variant, GlowQ-S, that applies the cached shared module only where it provides the largest benefit. Compared with strong baselines, our approach reduces TTFB by (5.6%) and increases throughput by (9.6%) on average, while reducing perplexity on WikiText-2 by (0.17%) and increasing downstream accuracy by 0.42 percentage points. The selective model GlowQ-S further reduces latency, cutting TTFB by (23.4%) and increasing throughput by (37.4%), while maintaining accuracy within 0.2 percentage points on average.

A faithful description of chemical processes requires exploring extended regions of the molecular potential energy surface (PES), which remains challenging for strongly correlated systems. Transferable deep-learning variational Monte Carlo (VMC) offers a promising route by efficiently solving the electronic Schrödinger equation jointly across molecular geometries at consistently high accuracy, yet its stochastic nature renders direct exploration of molecular configuration space nontrivial. Here, we present a framework for highly accurate ab initio exploration of PESs that combines transferable deep-learning VMC with a cost-effective estimation of energies, forces, and Hessians. By continuously sampling nuclear configurations during VMC optimization of electronic wave functions, we obtain transferable descriptions that achieve zero-shot chemical accuracy within chemically relevant distributions of molecular geometries. Throughout the subsequent characterization of molecular configuration space, the PES is evaluated only sparsely, with local approximations constructed by estimating VMC energies and forces at sampled geometries and aggregating the resulting noisy data using Gaussian process regression. Our method enables accurate and efficient exploration of complex PES landscapes, including structure relaxation, transition-state searches, and minimum-energy pathways, for both ground and excited states. This opens the door to studying bond breaking, formation, and large structural rearrangements in systems with pronounced multi-reference character.

Does structured intent representation generalize across languages and models? We study PPS (Prompt Protocol Specification), a 5W3H-based framework for structured intent representation in human-AI interaction, and extend prior Chinese-only evidence along three dimensions: two additional languages (English and Japanese), a fourth condition in which a user's simple prompt is automatically expanded into a full 5W3H specification by an AI-assisted authoring interface, and a new research question on cross-model output consistency. Across 2,160 model outputs (3 languages x 4 conditions x 3 LLMs x 60 tasks), we find that AI-expanded 5W3H prompts (Condition D) show no statistically significant difference in goal alignment from manually crafted 5W3H prompts (Condition C) across all three languages, while requiring only a single-sentence input from the user. Structured PPS conditions often reduce or reshape cross-model output variance, though this effect is not uniform across languages and metrics; the strongest evidence comes from identifying spurious low variance in unconstrained baselines. We also show that unstructured prompts exhibit a systematic dual-inflation bias: artificially high composite scores and artificially low apparent cross-model variance. These findings suggest that structured 5W3H representations can improve intent alignment and accessibility across languages and models, especially when AI-assisted authoring lowers the barrier for non-expert users.

RAG typically assumes centralized access to documents, which breaks down when knowledge is distributed across private data silos. We propose a secure Federated RAG system built using Flower that performs local silo retrieval, while server-side aggregation and text generation run inside an attested, confidential compute environment, enabling confidential remote LLM inference even in the presence of honest-but-curious or compromised servers. We also propose a cascading inference approach that incorporates a non-confidential third-party model (e.g., Amazon Nova) as auxiliary context without weakening confidentiality.

Local curvature of potential energy surfaces is critical for predicting certain experimental observables of molecules and materials from first principles, yet it remains far beyond reach for complex systems. In this work, we introduce a Hessian-informed Machine Learning Interatomic Potential (Hi-MLIP) that captures such curvature reliably, thereby enabling accurate analysis of associated thermodynamic and kinetic phenomena. To make Hessian supervision practically viable, we develop a highly efficient training protocol, termed Hessian INformed Training (HINT), achieving two to four orders of magnitude reduction for the requirement of expensive Hessian labels. HINT integrates critical techniques, including Hessian pre-training, configuration sampling, curriculum learning and stochastic projection Hessian loss. Enabled by HINT, Hi-MLIP significantly improves transition-state search and brings Gibbs free-energy predictions close to chemical accuracy especially in data-scarce regimes. Our framework also enables accurate treatment of strongly anharmonic hydrides, reproducing phonon renormalization and superconducting critical temperatures in close agreement with experiment while bypassing the computational bottleneck of anharmonic calculations. These results establish a practical route to enhancing curvature awareness of machine learning interatomic potentials, bridging simulation and experimental observables across a wide range of systems.

Probabilistic forecasting provides a principled framework for uncertainty quantification in dynamical systems by representing predictions as probability distributions rather than deterministic trajectories. However, existing forecasting approaches, whether physics-based or neural-network-based, remain fundamentally trajectory-oriented: predictive distributions are usually accessed through ensembles or sampling, rather than evolved directly as dynamical objects. A distribution-to-distribution (D2D) neural probabilistic forecasting framework is developed to operate directly on predictive distributions. The framework introduces a distributional encoding and decoding structure around a replaceable neural forecasting module, using kernel mean embeddings to represent input distributions and mixture density networks to parameterise output predictive distributions. This design enables recursive propagation of predictive uncertainty within a unified end-to-end neural architecture, with model training and evaluation carried out directly in terms of probabilistic forecast skill. The framework is demonstrated on the Lorenz63 chaotic dynamical system. Results show that the D2D model captures nontrivial distributional evolution under nonlinear dynamics, produces skillful probabilistic forecasts without explicit ensemble simulation, and remains competitive with, and in some cases outperforms, a simplified perfect model benchmark. These findings point to a new paradigm for probabilistic forecasting, in which predictive distributions are learned and evolved directly rather than reconstructed indirectly through ensemble-based uncertainty propagation.

Autonomous object search is challenging for mobile robots operating in indoor environments due to partial observability, perceptual uncertainty, and the need to trade off exploration and navigation efficiency. Classical probabilistic approaches explicitly represent uncertainty but typically rely on handcrafted action-selection heuristics, while deep reinforcement learning enables adaptive policies but often suffers from slow convergence and limited interpretability. This paper proposes a hybrid object-search framework that integrates Bayesian inference with deep reinforcement learning. The method maintains a spatial belief map over target locations, updated online through Bayesian inference from calibrated object detections, and trains a reinforcement learning policy to select navigation actions directly from this probabilistic representation. The approach is evaluated in realistic indoor simulation using Habitat 3.0 and compared against developed baseline strategies. Across two indoor environments, the proposed method improves success rate while reducing search effort. Overall, the results support the value of combining Bayesian belief estimation with learned action selection to achieve more efficient and reliable objectsearch behavior under partial observability.

Arithmetic puzzle games provide a controlled setting for studying difficulty in mathematical reasoning tasks, a core challenge in adaptive learning systems. We investigate the structural determinants of difficulty in a class of integer arithmetic puzzles inspired by number games. We formalize the problem and develop an exact dynamic-programming solver that enumerates reachable targets, extracts minimal-operation witnesses, and enables large-scale labeling. Using this solver, we construct a dataset of over 3.4 million instances and define difficulty via the minimum number of operations required to reach a target. We analyze the relationship between difficulty and solver-derived features. While baseline machine learning models based on bag- and target-level statistics can partially predict solvability, they fail to reliably distinguish easy instances. In contrast, we show that difficulty is fully determined by a small set of interpretable structural attributes derived from exact witnesses. In particular, the number of input values used in a minimal construction serves as a minimal sufficient statistic for difficulty under this labeling. These results provide a transparent, computationally grounded account of puzzle difficulty that bridges symbolic reasoning and data-driven modeling. The framework supports explainable difficulty estimation and principled task sequencing, with direct implications for adaptive arithmetic learning and intelligent practice systems.

Automatic image rotation estimation is a key preprocessing step in many vision pipelines. This task is challenging because angles have circular topology, creating boundary discontinuities that hinder standard regression methods. We present a comprehensive study of five circular-aware methods for global orientation estimation: direct angle regression with circular loss, classification via angular binning, unit-vector regression, phase-shifting coder, and circular Gaussian distribution. Using transfer learning from ImageNet-pretrained models, we systematically evaluate these methods across sixteen modern architectures by adapting their output heads for rotation-specific predictions. Our results show that probabilistic methods, particularly the circular Gaussian distribution, are the most robust across architectures, while classification achieves the best accuracy on well-matched backbones but suffers training instabilities on others. The best configuration (classification with EfficientViT-B3) achieves a mean absolute error (MAE) of 1.23° (mean across five independent runs) on the DRC-D dataset, while the circular Gaussian distribution with MambaOut Base achieves a virtually identical 1.24° with greater robustness across backbones. Training and evaluating our top-performing method-architecture combinations on COCO 2014, the best configuration reaches 3.71° MAE, improving substantially over prior work, with further improvement to 2.84° on the larger COCO 2017 dataset.

Although deep research agents (DRAs) have emerged as a promising paradigm for complex information synthesis, their evaluation remains constrained by ad hoc empirical benchmarks. These heuristic approaches do not rigorously model agent behavior or adequately stress-test long-horizon synthesis and ambiguity resolution. To bridge this gap, we formalize DRA behavior through the lens of category theory, modeling deep research workflow as a composition of structure-preserving maps (functors). Grounded in this theoretical framework, we introduce a novel mechanism-aware benchmark with 296 questions designed to stress-test agents along four interpretable axes: traversing sequential connectivity chains, verifying intersections within V-structure pullbacks, imposing topological ordering on retrieved substructures, and performing ontological falsification via the Yoneda Probe. Our rigorous evaluation of 11 leading models establishes a persistently low baseline, with the state-of-the-art achieving only a 19.9\% average accuracy, exposing the difficulty of formal structural stress-testing. Furthermore, our findings reveal a stark dichotomy in the current AI capabilities. While advanced deep research pipelines successfully redefine dynamic topological re-ordering and exhibit robust ontological verification -- matching pure reasoning models in falsifying hallucinated premises -- they almost universally collapse on multi-hop structural synthesis. Crucially, massive performance variance across tasks exposes a lingering reliance on brittle heuristics rather than a systemic understanding. Ultimately, this work demonstrates that while top-tier autonomous agents can now organically unify search and reasoning, achieving a generalized mastery over complex structural information remains a formidable open challenge.\footnote{Our implementation will be available at https://github.com/tzq1999/CDR.

In this paper, we establish the novel insight that an off-the-shelf LLM can function as an excellent token compressor and decompressor. To demonstrate, we design a self-expressive autoencoding learning framework fine-tunes a pretrained LLM to translate long texts into a compact internal language of discrete, variable-length latent codes, termed Z-tokens, and to reconstruct the original text exactly from them. The resulting representation is content-adaptive: semantically dense segments receive more Z-tokens, while redundant or predictable regions are aggressively compressed, via lightweight LoRA-based adapter heads. Empirically, our method achieves up to 18 times token reduction on Wikipedia, CNN/DailyMail, HotpotQA, and Qulac-style long-query datasets, while preserving reconstruction fidelity and downstream performance. This simple yet effective design supports applications including prompt compression and autoregressive generation directly in the Z-token space, offering a potential pathway toward token-efficient long-context reasoning.

Distributed intelligence in industrial networks increasingly integrates sensing, communication, and computation across heterogeneous and resource constrained devices. Federated learning (FL) enables collaborative model training in such environments, but its reliability is affected by inconsistent client behaviour, noisy sensing conditions, and the presence of faulty or adversarial updates. Trust based mechanisms are commonly used to mitigate these effects, yet most remain statistical and heuristic, relying on fixed parameters or simple adaptive rules that struggle to accommodate changing operating conditions. This paper presents a lightweight agentic trust coordination approach for FL in sustainable and resilient industrial networks. The proposed Agentic Trust Control Layer operates as a server side control loop that observes trust related and system level signals, interprets their evolution over time, and applies targeted trust adjustments when instability is detected. The approach extends prior adaptive trust mechanisms by enabling context aware intervention decisions, rather than relying on fixed or purely reactive parameter updates. By explicitly separating observation, reasoning, and action, the proposed framework supports stable FL operation without modifying client side training or increasing communication overhead.

The effectiveness of Retrieval-Augmented Generation (RAG) is highly dependent on how documents are chunked, that is, segmented into smaller units for indexing and retrieval. Yet, commonly used "one-size-fits-all" approaches often fail to capture the nuanced structure and semantics of diverse texts. Despite its central role, chunking lacks a dedicated evaluation framework, making it difficult to assess and compare strategies independently of downstream performance. We challenge this paradigm by introducing Adaptive Chunking, a framework that selects the most suitable chunking strategy for each document based on a set of five novel intrinsic, document-based metrics: References Completeness (RC), Intrachunk Cohesion (ICC), Document Contextual Coherence (DCC), Block Integrity (BI), and Size Compliance (SC), which directly assess chunking quality across key dimensions. To support this framework, we also introduce two new chunkers, an LLM-regex splitter and a split-then-merge recursive splitter, alongside targeted post-processing techniques. On a diverse corpus spanning legal, technical, and social science domains, our metric-guided adaptive method significantly improves downstream RAG performance. Without changing models or prompts, our framework increases RAG outcomes, raising answers correctness to 72% (from 62-64%) and increasing the number of successfully answered questions by over 30% (65 vs. 49). These results demonstrate that adaptive, document-aware chunking, guided by a complementary suite of intrinsic metrics, offers a practical and effective path to more robust RAG systems. Code available at https://github.com/ekimetrics/adaptive-chunking.

As generative AI tools like ChatGPT enter classrooms, workplaces and everyday thinking, writing is at risk of becoming a formality -- outsourced, automated and stripped of its cognitive value. But writing is not just output; it is how we learn to think. This paper explores what is lost when we let machines write for us, drawing on cognitive psychology, educational theory and real classroom practices. We argue that the process of writing -- messy, slow, often frustrating -- is where a human deep learning happens. The paper also explores the current possibilities of AI-text detection, how educators can adapt through smarter pedagogy rather than bans, and why the ability to recognize machine-generated language may become a critical literacy of the 21st century. In a world where writing can be faked, learning can not.

Automated Vehicle (AV) control in mixed traffic, where AVs coexist with human-driven vehicles, poses significant challenges in balancing safety, efficiency, comfort, fuel efficiency, and compliance with traffic rules while capturing heterogeneous driver behavior. Traditional car-following models, such as the Intelligent Driver Model (IDM), often struggle to generalize across diverse traffic scenarios and typically do not account for fuel efficiency, motivating the use of learning-based approaches. Although Deep Reinforcement Learning (DRL) has shown strong microscopic performance in car-following conditions, its macroscopic traffic flow characteristics remain underexplored. This study focuses on analyzing the macroscopic traffic flow characteristics and fuel efficiency of DRL-based models in mixed traffic. A Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is implemented for AVs' control and trained using the NGSIM highway dataset, enabling realistic interaction with human-driven vehicles. Traffic performance is evaluated using the Fundamental Diagram (FD) under varying driver heterogeneity, heterogeneous time-gap penetration levels, and different shares of RL-controlled vehicles. A macroscopic level comparison of fuel efficiency between the RL-based AV model and the IDM is also conducted. Results show that traffic performance is sensitive to the distribution of safe time gaps and the proportion of RL vehicles. Transitioning from fully human-driven to fully RL-controlled traffic can increase road capacity by approximately 7.52%. Further, RL-based AVs also improve average fuel efficiency by about 28.98% at higher speeds (above 50 km/h), and by 1.86% at lower speeds (below 50 km/h) compared to the IDM. Overall, the DRL framework enhances traffic capacity and fuel efficiency without compromising safety.

Interest in the concept of AI-driven harmful manipulation is growing, yet current approaches to evaluating it are limited. This paper introduces a framework for evaluating harmful AI manipulation via context-specific human-AI interaction studies. We illustrate the utility of this framework by assessing an AI model with 10,101 participants spanning interactions in three AI use domains (public policy, finance, and health) and three locales (US, UK, and India). Overall, we find that that the tested model can produce manipulative behaviours when prompted to do so and, in experimental settings, is able to induce belief and behaviour changes in study participants. We further find that context matters: AI manipulation differs between domains, suggesting that it needs to be evaluated in the high-stakes context(s) in which an AI system is likely to be used. We also identify significant differences across our tested geographies, suggesting that AI manipulation results from one geographic region may not generalise to others. Finally, we find that the frequency of manipulative behaviours (propensity) of an AI model is not consistently predictive of the likelihood of manipulative success (efficacy), underscoring the importance of studying these dimensions separately. To facilitate adoption of our evaluation framework, we detail our testing protocols and make relevant materials publicly available. We conclude by discussing open challenges in evaluating harmful manipulation by AI models.

Weight pruning is a standard technique for compressing large language models, yet its effect on learned internal representations remains poorly understood. We present the first systematic study of how unstructured pruning reshapes the feature geometry of language models, using Sparse Autoencoders (SAEs) as interpretability probes. Across three model families (Gemma 3 1B, Gemma 2 2B, Llama 3.2 1B), two pruning methods (magnitude and Wanda), and six sparsity levels (0--60%), we investigate five research questions spanning seed stability, feature survival, SAE transferability, feature fragility, and causal relevance. Our most striking finding is that rare SAE features--those with low firing rates--survive pruning far better than frequent ones, with within-condition Spearman correlations of rho = -1.0 in 11 of 17 experimental conditions. This counter-intuitive result suggests that pruning acts as implicit feature selection, preferentially destroying high-frequency generic features while preserving specialized rare ones. We further show that Wanda pruning preserves feature structure up to 3.7x better than magnitude pruning, that pre-trained SAEs remain viable on Wanda-pruned models up to 50% sparsity, and that geometric feature survival does not predict causal importance--a dissociation with implications for interpretability under compression.

Alzheimer's disease (AD) is a growing global health challenge as populations age, and timely, accurate diagnosis is essential to reduce individual and societal burden. However, real-world AD assessment is hampered by incomplete, heterogeneous multimodal data and variability across sites and patient demographics. Although large language models (LLMs) have shown promise in biomedicine, their use in AD has largely been confined to answering narrow, disease-specific questions rather than generating comprehensive diagnostic reports that support clinical decision-making. Here we expand LLM capabilities for clinical decision support by introducing AD-CARE, a modality-agnostic agent that performs guideline-grounded diagnostic assessment from incomplete, heterogeneous inputs without imputing missing modalities. By dynamically orchestrating specialized diagnostic tools and embedding clinical guidelines into LLM-driven reasoning, AD-CARE generates transparent, report-style outputs aligned with real-world clinical workflows. Across six cohorts comprising 10,303 cases, AD-CARE achieved 84.9% diagnostic accuracy, delivering 4.2%-13.7% relative improvements over baseline methods. Despite cohort-level differences, dataset-specific accuracies remain robust (80.4%-98.8%), and the agent consistently outperforms all baselines. AD-CARE reduced performance disparities across racial and age subgroups, decreasing the average dispersion of four metrics by 21%-68% and 28%-51%, respectively. In a controlled reader study, the agent improved neurologist and radiologist accuracy by 6%-11% and more than halved decision time. The framework yielded 2.29%-10.66% absolute gains over eight backbone LLMs and converges their performance. These results show that AD-CARE is a scalable, practically deployable framework that can be integrated into routine clinical workflows for multimodal decision support in AD.

Efficient global optimization (EGO) is one of the most widely used noise-free Bayesian optimization algorithms.It comprises the Gaussian process (GP) surrogate model and expected improvement (EI) acquisition function. In practice, when EGO is applied, a scalar matrix of a small positive value (also called a nugget or jitter) is usually added to the covariance matrix of the deterministic GP to improve numerical stability. We refer to this EGO with a positive nugget as the practical EGO. Despite its wide adoption and empirical success, to date, cumulative regret bounds for practical EGO have yet to be established. In this paper, we present for the first time the cumulative regret upper bound of practical EGO. In particular, we show that practical EGO has sublinear cumulative regret bounds and thus is a no-regret algorithm for commonly used kernels including the squared exponential (SE) and Matérn kernels ($ν>\frac{1}{2}$). Moreover, we analyze the effect of the nugget on the regret bound and discuss the theoretical implication on its choice. Numerical experiments are conducted to support and validate our findings.

Current Large Language Models (LLMs) mostly use BPE (Byte Pair Encoding) based tokenizers, which are very effective for simple structured Latin scripts such as English. However, standard BPE tokenizers struggle to process complex Abugida scripts due to their structural complexity. The problem is that these tokenizers break complex conjuncts, which are multi-codepoint grapheme clusters, into meaningless sub-character units. This degrades the LLM's reasoning efficiency by forcing it to learn basic orthographic structures at inference time and raises inference costs, resulting in a significant "Token Tax" for the Global South. We propose a new three-layer architecture, the WWHO (Where-What-How Often), and an algorithm named SGPE (Syllable-aware Grapheme Pair Encoding) that separates the linguistic rules of the script from the statistical compression process while enabling seamless multilingual tokenization. Using Sinhala and Devanagari (Hindi/Sanskrit) as highly complex Abugida scripts, we trained WWHO on a cleaned 30-million-sentence dataset and evaluated on a 1,499,950-sentence test set. For Sinhala, SGPE achieves a Token to Word Ratio (TWR) of 1.274 with 4.83 characters per token, representing a 61.7 percent reduction in tokens compared to OpenAI's o200k base. For Hindi, it achieves a TWR of 1.181 (27.0 percent reduction vs o200k). On the mixed-script (Sinhala, Devanagari, and English) dataset, SGPE achieves an overall TWR of 1.240, representing token reductions of 36.7 percent, 39.6 percent, and 60.2 percent relative to o200k base, Llama 4 Scout, and DeepSeek V3, respectively. This effectively extends the usable context window by up to 4.38 times for these Abugida languages while ensuring a Linguistic Zero-Breakage Guarantee, which ensures that no valid syllable is ever split across multiple tokens.

Directed Acyclic Graphs (DAGs) are widely used to represent structured knowledge in scientific and technical domains. However, datasets for real-world DAGs remain scarce because constructing them typically requires expert interpretation of domain documents. We study Doc2SemDAG construction: recovering a preferred semantic DAG from a document together with the cited evidence and context that explain it. This problem is challenging because a document may admit multiple plausible abstractions, the intended structure is often implicit, and the supporting evidence is scattered across prose, equations, captions, and figures. To address these challenges, we leverage scientific papers containing explicit DAG figures as a natural source of supervision. In this setting, the DAG figure provides the DAG structure, while the accompanying text provides context and explanation. We introduce DAGverse, a framework for constructing document-grounded semantic DAGs from online scientific papers. Its core component, DAGverse-Pipeline, is a semi-automatic system designed to produce high-precision semantic DAG examples through figure classification, graph reconstruction, semantic grounding, and validation. As a case study, we test the framework for causal DAGs and release DAGverse-1, a dataset of 108 expert-validated semantic DAGs with graph-level, node-level, and edge-level evidence. Experiments show that DAGverse-Pipeline outperforms existing Vision-Language Models on DAG classification and annotation. DAGverse provides a foundation for document-grounded DAG benchmarks and opens new directions for studying structured reasoning grounded in real-world evidence.

Federated Learning (FL) is a paradigm for training machine learning (ML) models in collaborative settings while preserving participants' privacy by keeping raw data local. A key requirement for the use of FL in production is reliability, as insufficient reliability can compromise the validity, stability, and reproducibility of learning outcomes. FL inherently operates as a distributed system and is therefore susceptible to crash failures, network partitioning, and other fault scenarios. Despite this, the impact of such failures on FL outcomes has not yet been studied systematically. In this paper, we address this gap by investigating the impact of missing participants in FL. To this end, we conduct extensive experiments on image, tabular, and time-series data and analyze how the absence of participants affects model performance, taking into account influencing factors such as data skewness, different availability patterns, and model architectures. Furthermore, we examine scenario-specific aspects, including the utility of the global model for missing participants. Our experiments provide detailed insights into the effects of various influencing factors. In particular, we show that data skewness has a strong impact, often leading to overly optimistic model evaluations and, in some cases, even altering the effects of other influencing factors.

Low-altitude communications can promote the integration of aerial and terrestrial wireless resources, expand network coverage, and enhance transmission quality, thereby empowering the development of sixth-generation (6G) mobile communications. As an enabler for low-altitude transmission, 3D channel fingerprints (3D-CF), also referred to as the 3D radio map or 3D channel knowledge map, are expected to enhance the understanding of communication environments and assist in the acquisition of channel state information (CSI), thereby avoiding repeated estimations and reducing computational complexity. In this paper, we propose a modularized multimodal framework to construct 3D-CF. Specifically, we first establish the 3D-CF model as a collection of CSI-tuples based on Rician fading channels, with each tuple comprising the low-altitude vehicle's (LAV) positions and its corresponding statistical CSI. In consideration of the heterogeneous structures of different prior data, we formulate the 3D-CF construction problem as a multimodal regression task, where the target channel information in the CSI-tuple can be estimated directly by its corresponding LAV positions, together with communication measurements and geographic environment maps. Then, a high-efficiency multimodal framework is proposed accordingly, which includes a correlation-based multimodal fusion (Corr-MMF) module, a multimodal representation (MMR) module, and a CSI regression (CSI-R) module. Numerical results show that our proposed framework can efficiently construct 3D-CF and achieve at least 27.5% higher accuracy than the state-of-the-art algorithms under different communication scenarios, demonstrating its competitive performance and excellent generalization ability. We also analyze the computational complexity and illustrate its superiority in terms of the inference time.

In this paper, we address post-training quantization (PTQ) for large language models (LLMs) from an overlooked perspective: given a pre-trained high-precision LLM, the predominant sequential quantization framework treats different layers equally, but this may be not optimal in challenging bit-width settings. We empirically study the quantization impact of different layers on model accuracy, and observe that: (1) shallow/deep layers are usually more sensitive to quantization than intermediate layers; (2) among shallow/deep layers, the most sensitive one is the first/last layer, which exhibits significantly larger quantization error than others. These empirical observations imply that the quantization design for different layers of LLMs is required on multiple levels instead of a single level shared to all layers. Motivated by this, we propose a new PTQ framework termed Sliding-layer Quantization (SliderQuant) that relies on a simple adaptive sliding quantization concept facilitated by few learnable parameters. The base component of SliderQuant is called inter-layer sliding quantization, which incorporates three types of novel sliding window designs tailored for addressing the varying quantization sensitivity of shallow, intermediate and deep layers. The other component is called intra-layer sliding quantization that leverages an incremental strategy to quantize each window. As a result, SliderQuant has a strong ability to reduce quantization errors across layers. Extensive experiments on basic language generation, zero-shot commonsense reasoning and challenging math and code tasks with various LLMs, including Llama/Llama2/Llama3/Qwen2.5 model families, DeepSeek-R1 distilled models and large MoE models, show that our method outperforms existing PTQ methods (including the latest PTQ methods using rotation transformations) for both weight-only quantization and weight-activation quantization.

Gait is increasingly recognized as a vital sign, yet current approaches treat it as a symptom of specific pathologies rather than a systemic biomarker. We developed a gait foundation model for 3D skeletal motion from 3,414 deeply phenotyped adults, recorded via a depth camera during five motor tasks. Learned embeddings outperformed engineered features, predicting age (Pearson r = 0.69), BMI (r = 0.90), and visceral adipose tissue area (r = 0.82). Embeddings significantly predicted 1,980 of 3,210 phenotypic targets; after adjustment for age, BMI, VAT, and height, gait provided independent gains in all 18 body systems in males and 17 of 18 in females, and improved prediction of clinical diagnoses and medication use. Anatomical ablation revealed that legs dominated metabolic and frailty predictions while torso encoded sleep and lifestyle phenotypes. These findings establish gait as an independent multi-system biosignal, motivating translation to consumer-grade video and its integration as a scalable, passive vital sign.

The probabilistic abstract interpretation framework of neural network analysis analyzes a neural network by analyzing its density distribution flow of all possible inputs. The grids approximation is one of abstract domains the framework uses which abstracts concrete space into grids. In this paper, we introduce two novel approximation methods: distribution approximation and clusters approximation. We show how these two methods work in theory with corresponding abstract transformers with help of illustrations of some simple examples.

Hateful content online is often expressed using fact-like, not necessarily correct information, especially in coordinated online harassment campaigns and extremist propaganda. Failing to jointly address hate speech (HS) and misinformation can deepen prejudice, reinforce harmful stereotypes, and expose bystanders to psychological distress, while polluting public debate. Moreover, these messages require more effort from content moderators because they must assess both harmfulness and veracity, i.e., fact-check them. To address this challenge, we release WSF-ARG+, the first dataset which combines hate speech with check-worthiness information. We also introduce a novel LLM-in-the-loop framework to facilitate the annotation of check-worthy claims. We run our framework, testing it with 12 open-weight LLMs of different sizes and architectures. We validate it through extensive human evaluation, and show that our LLM-in-the-loop framework reduces human effort without compromising the annotation quality of the data. Finally, we show that HS messages with check-worthy claims show significantly higher harassment and hate, and that incorporating check-worthiness labels improves LLM-based HS detection up to 0.213 macro-F1 and to 0.154 macro-F1 on average for large models.

We introduce CRAFT, a multi-agent benchmark for evaluating pragmatic communication in large language models under strict partial information. In this setting, multiple agents with complementary but incomplete views must coordinate through natural language to construct a shared 3D structure that no single agent can fully observe. We formalize this problem as a multi-sender pragmatic reasoning task and provide a diagnostic framework that decomposes failures into spatial grounding, belief modeling and pragmatic communication errors, including a taxonomy of behavioral failure profiles in both frontier and open-weight models. Across a diverse set of models, including 8 open-weight and 7 frontier including reasoning models, we find that stronger reasoning ability does not reliably translate to better coordination: smaller open-weight models often match or outperform frontier systems, and improved individual communication does not guarantee successful collaboration. These results suggest that multi-agent coordination remains a fundamentally unsolved challenge for current language models. Our code can be found at https://github.com/csu-signal/CRAFT

Probabilistic abstract interpretation is a theory used to extract particular properties of a computer program when it is infeasible to test every single inputs. In this paper we apply the theory on neural networks for the same purpose: to analyse density distribution flow of all possible inputs of a neural network when a network has uncountably many or countable but infinitely many inputs. We show how this theoretical framework works in neural networks and then discuss different abstract domains and corresponding Moore-Penrose pseudo-inverses together with abstract transformers used in the framework. We also present experimental examples to show how this framework helps to analyse real world problems.

This paper investigates the susceptibility to model integrity attacks that overload virtual machines assigned by the k-means algorithm used for resource provisioning in fog networks. The considered k-means algorithm runs two phases iteratively: offline clustering to form clusters of requested workload and online classification of new incoming requests into offline-created clusters. First, we consider an evasion attack against the classifier in the online phase. A threat actor launches an exploratory attack using query-based reverse engineering to discover the Machine Learning (ML) model (the clustering scheme). Then, a passive causative (evasion) attack is triggered in the offline phase. To defend the model, we suggest a proactive method using adversarial training to introduce attack robustness into the classifier. Our results show that our mitigation technique effectively maintains the stability of the resource provisioning system against attacks.

Trajectory anomaly detection underpins applications from fraud detection to urban mobility analysis. Dense GPS methods preserve fine-grained evidence such as abnormal speeds and short-duration events, but their quadratic cost makes multi-month analysis intractable; consequently, no existing approach detects anomalies over multi-month dense GPS trajectories. The field instead relies on scalable sparse stay-point methods that discard this evidence, forcing separate architectures for each regime and preventing knowledge transfer. We argue this bottleneck is unnecessary: human trajectories, dense or sparse, share a natural two-dimensional cyclic structure along within-day and across-day axes. We therefore propose TITAnD (Trajectory Image Transformer for Anomaly Detection), which reformulates trajectory anomaly detection as a vision problem by representing trajectories as a Hyperspectral Trajectory Image (HTI): a day x time-of-day grid whose channels encode spatial, semantic, temporal, and kinematic information from either modality, unifying both under a single representation. Under this formulation, agent-level detection reduces to image classification and temporal localization to semantic segmentation. To model this representation, we introduce the Cyclic Factorized Transformer (CFT), which factorizes attention along the two temporal axes, encoding the cyclic inductive bias of human routines, while reducing attention cost by orders of magnitude and enabling dense multi-month anomaly detection for the first time. Empirically, TITAnD achieves the best AUC-PR across sparse and dense benchmarks, surpassing vision models like UNet while being 11-75x faster than the Transformer with comparable memory, demonstrating that vision reformulation and structure-aware modeling are jointly essential. Code will be made public soon.

Large language models (LLMs) hold considerable potential for advancing scientific discovery, yet systematic assessment of their dynamic reasoning in real-world research remains limited. Current scientific evaluation benchmarks predominantly rely on static, single-turn Question Answering (QA) formats, which are inadequate for measuring model performance in complex scientific tasks that require multi-step iteration and experimental interaction. To address this gap, we introduce MolQuest, a novel agent-based evaluation framework for molecular structure elucidation built upon authentic chemical experimental data. Unlike existing datasets, MolQuest formalizes molecular structure elucidation as a multi-turn interactive task, requiring models to proactively plan experimental steps, integrate heterogeneous spectral sources (e.g., NMR, MS), and iteratively refine structural hypotheses. This framework systematically evaluates LLMs' abductive reasoning and strategic decision-making abilities within a vast and complex chemical space. Empirical results reveal that contemporary frontier models exhibit significant limitations in authentic scientific scenarios: notably, even state-of-the-art (SOTA) models achieve an accuracy of only approximately 50%, while the performance of most other models remains below the 30% threshold. This work provides a reproducible and extensible framework for science-oriented LLM evaluation, our findings highlight the critical gap in current LLMs' strategic scientific reasoning, setting a clear direction for future research toward AI that can actively participate in the scientific process.

Explainable AI (XAI) methods are commonly evaluated with functional metrics such as correctness, which computationally estimate how accurately an explanation reflects the model's reasoning. Higher correctness is assumed to produce better human understanding, but this link has not been tested experimentally with controlled levels. We conducted a user study (N=200) that manipulated explanation correctness at four levels (100%, 85%, 70%, 55%) in a time series classification task where participants could not rely on domain knowledge or visual intuition and instead predicted the AI's decisions based on explanations (forward simulation). Correctness affected understanding, but not at every level: performance dropped at 70% and 55% correctness relative to fully correct explanations, while further degradation below 70% produced no additional loss. Rather than shifting performance uniformly, lower correctness decreased the proportion of participants who learned the decision pattern. At the same time, even fully correct explanations did not guarantee understanding, as only a subset of participants achieved high accuracy. Exploratory analyses showed that self-reported ratings correlated with demonstrated performance only when explanations were fully correct and participants had learned the pattern. These findings show that not all differences in functional correctness translate to differences in human understanding, underscoring the need to validate functional metrics against human outcomes.

Out-of-distribution (OOD) detection aims to identify samples that deviate from in-distribution (ID). One popular pipeline addresses this by introducing negative labels distant from ID classes and detecting OOD based on their distance to these labels. However, such labels may present poor activation on OOD samples, failing to capture the OOD characteristics. To address this, we propose \underline{T}est-time \underline{A}ctivated \underline{N}egative \underline{L}abels (TANL) by dynamically evaluating activation levels across the corpus dataset and mining candidate labels with high activation responses during the testing process. Specifically, TANL identifies high-confidence test images online and accumulates their assignment probabilities over the corpus to construct a label activation metric. Such a metric leverages historical test samples to adaptively align with the test distribution, enabling the selection of distribution-adaptive activated negative labels. By further exploring the activation information within the current testing batch, we introduce a more fine-grained, batch-adaptive variant. To fully utilize label activation knowledge, we propose an activation-aware score function that emphasizes negative labels with stronger activations, boosting performance and enhancing its robustness to the label number. Our TANL is training-free, test-efficient, and grounded in theoretical justification. Experiments on diverse backbones and wide task settings validate its effectiveness. Notably, on the large-scale ImageNet benchmark, TANL significantly reduces the FPR95 from 17.5\% to 9.8\%. Codes are available at \href{https://github.com/YBZh/OpenOOD-VLM}{YBZh/OpenOOD-VLM}.

Spatial Transcriptomics (ST) provides spatially-resolved gene expression, offering crucial insights into tissue architecture and complex diseases. However, its prohibitive cost limits widespread adoption, leading to significant attention on inferring spatial gene expression from readily available whole slide images. While graph neural networks have been proposed to model interactions between tissue regions, their reliance on pre-defined sparse graphs prevents them from considering potentially interacting spot pairs, resulting in a structural limitation in capturing complex biological relationships. To address this, we propose FEAST (Fully connected Expressive Attention for Spatial Transcriptomics), an attention-based framework that models the tissue as a fully connected graph, enabling the consideration of all pairwise interactions. To better reflect biological interactions, we introduce negative-aware attention, which models both excitatory and inhibitory interactions, capturing essential negative relationships that standard attention often overlooks. Furthermore, to mitigate the information loss from truncated or ignored context in standard spot image extraction, we introduce an off-grid sampling strategy that gathers additional images from intermediate regions, allowing the model to capture a richer morphological context. Experiments on public ST datasets show that FEAST surpasses state-of-the-art methods in gene expression prediction while providing biologically plausible attention maps that clarify positive and negative interactions. Our code is available at https://github.com/starforTJ/ FEAST.

Large language models and autonomous agents are increasingly explored for EDA automation, but many existing integrations still rely on script-level or request-level interactions, which makes it difficult to preserve tool state and support iterative optimization in real production-oriented environments. In this work, we present FluxEDA, a unified and stateful infrastructure substrate for agentic EDA. FluxEDA introduces a managed gateway-based execution interface with structured request and response handling. It also maintains persistent backend instances. Together, these features allow upper-layer agents and programmable clients to interact with heterogeneous EDA tools through preserved runtime state, rather than through isolated shell invocations. We evaluate the framework using two representative commercial backend case studies: automated post-route timing ECO and standard-cell sub-library optimization. The results show that FluxEDA can support multi-step analysis and optimization over real tool contexts, including state reuse, rollback, and coordinated iterative execution. These findings suggest that a stateful and governed infrastructure layer is a practical foundation for agent-assisted EDA automation.

Combinatorial optimization problems like the Traveling Salesman Problem are critical in industry yet NP-hard. Neural Combinatorial Optimization has shown promise, but its reliance on online reinforcement learning (RL) hampers deployment and underutilizes decades of algorithmic knowledge. We address these limitations by applying the offline RL framework, Decision Transformer, to learn superior strategies directly from datasets of heuristic solutions; it aims to not only to imitate but to synthesize and outperform them. Concretely, we (i) integrate a Pointer Network to handle the instance-dependent, variable action space of node selection, and (ii) employ expectile regression for optimistic conditioning of Return-to-Go, which is crucial for instances with widely varying optimal values. Experiments show that our method consistently produces higher-quality tours than the four classical heuristics it is trained on, demonstrating the potential of offline RL to unlock and exceed the performance embedded in existing domain knowledge.

Skin diseases are a major public health concern worldwide, and their detection is often challenging without access to dermatological expertise. In countries like Bangladesh, which is highly populated, the number of qualified skin specialists and diagnostic instruments is insufficient to meet the demand. Due to the lack of proper detection and treatment of skin diseases, that may lead to severe health consequences including death. Common properties of skin diseases are, changing the color, texture, and pattern of skin and in this era of artificial intelligence and machine learning, we are able to detect skin diseases by using image processing and computer vision techniques. In response to this challenge, we develop a publicly available dataset focused on common skin disease detection using machine learning techniques. We focus on five prevalent skin diseases in Bangladesh: Contact Dermatitis, Vitiligo, Eczema, Scabies, and Tinea Ringworm. The dataset consists of 1612 images (of which, 250 are distinct while others are augmented), collected directly from patients at the outpatient department of Faridpur Medical College, Faridpur, Bangladesh. The data comprises of 302, 381, 301, 316, and 312 images of Dermatitis, Eczema, Scabies, Tinea Ringworm, and Vitiligo, respectively. Although the data are collected regionally, the selected diseases are common across many countries especially in South Asia, making the dataset potentially valuable for global applications in machine learning-based dermatology. We also apply several machine learning and deep learning models on the dataset and report classification performance. We expect that this research would garner attention from machine learning and deep learning researchers and practitioners working in the field of automated disease diagnosis.

This study compares the impact of natural and synthetic data on training and evaluating large language models (LLMs), using the case of passive verb alternation in French and Italian. We use Blackbird Language Matrices (BLMs), structured datasets designed to probe linguistic knowledge of underlying patterns across sentence sets. We compare structured templates instantiated with natural sentences extracted from Universal Dependencies to structured templates of synthetic sentences. Experiments show that while models achieve ceiling performance when trained and tested on synthetic datasets, they do not reliably generalize to natural sentences. In contrast, models trained on natural data exhibit robust performance across both natural and synthetic test suites, demonstrating their superior ability to capture abstract linguistic patterns. These results corroborate the value of natural data and of structured set ups in linguistic evaluation for probing LLMs' syntactic and semantic knowledge.

The emergence of Large Language Models (LLMs) has catalyzed a paradigm shift in programming, giving rise to "vibe coding", where users can build complete projects and even control computers using natural language instructions. This paradigm has driven automated webpage development, but it introduces a new requirement about how to automatically verify whether the web functionalities are reliably implemented. Existing works struggle to adapt, relying on static visual similarity or predefined checklists that constrain their utility in open-ended environments. Furthermore, they overlook a vital aspect of software quality, namely latent logical constraints. To address these gaps, we introduce WebTestBench, a benchmark for evaluating end-to-end automated web testing. WebTestBench encompasses comprehensive dimensions across diverse web application categories. We decompose the testing process into two cascaded sub-tasks, checklist generation and defect detection, and propose WebTester, a baseline framework for this task. Evaluating popular LLMs with WebTester reveals severe challenges, including insufficient test completeness, detection bottlenecks, and long-horizon interaction unreliability. These findings expose a substantial gap between current computer-use agent capabilities and industrial-grade deployment demands. We hope that WebTestBench provides valuable insights and guidance for advancing end-to-end automated web testing. Our dataset and code are available at https://github.com/friedrichor/WebTestBench.

Demographic parity (DP) is a widely used group fairness criterion requiring predictive distributions to be invariant across sensitive groups. While natural in classification, full distributional DP is often overly restrictive in regression and can lead to substantial accuracy loss. We propose a relaxation of DP tailored to regression, enforcing parity only at a finite set of quantile levels and/or score thresholds. Concretely, we introduce a novel (${\ell}$, Z)-fair predictor, which imposes groupwise CDF constraints of the form F f |S=s (z m ) = ${\ell}$ m for prescribed pairs (${\ell}$ m , z m ). For this setting, we derive closed-form characterizations of the optimal fair discretized predictor via a Lagrangian dual formulation and quantify the discretization cost, showing that the risk gap to the continuous optimum vanishes as the grid is refined. We further develop a model-agnostic post-processing algorithm based on two samples (labeled for learning a base regressor and unlabeled for calibration), and establish finite-sample guarantees on constraint violation and excess penalized risk. In addition, we introduce two alternative frameworks where we match group and marginal CDF values at selected score thresholds. In both settings, we provide closed-form solutions for the optimal fair discretized predictor. Experiments on synthetic and real datasets illustrate an interpretable fairness-accuracy trade-off, enabling targeted corrections at decision-relevant quantiles or thresholds while preserving predictive performance.

The landscape of extremely low-resource machine translation (MT) is characterized by perplexing variability in reported performance, often making results across different language pairs difficult to contextualize. For researchers focused on specific language groups -- such as ancient languages -- it is nearly impossible to determine if breakthroughs reported in other contexts (e.g., native African or American languages) result from superior methodologies or are merely artifacts of benchmark collection. To address this problem, we introduce the FRED Difficulty Metrics, which include the Fertility Ratio (F), Retrieval Proxy (R), Pre-training Exposure (E), and Corpus Diversity (D) and serve as dataset-intrinsic metrics to contextualize reported scores. These metrics reveal that a significant portion of result variability is explained by train-test overlap and pre-training exposure rather than model capability. Additionally, we identify that some languages -- particularly extinct and non-Latin indigenous languages -- suffer from poor tokenization coverage (high token fertility), highlighting a fundamental limitation of transferring models from high-resource languages that lack a shared vocabulary. By providing these indices alongside performance scores, we enable more transparent evaluation of cross-lingual transfer and provide a more reliable foundation for the XLR MT community.

Robust Support Vector Machines (R-SVMs) address feature noise by adopting a worst-case robust formulation that explicitly incorporates uncertainty sets into training. While this robustness improves reliability, it also leads to increased computational cost. In this work, we develop safe sample screening rules for R-SVMs that reduce the training complexity without affecting the optimal solution. To the best of our knowledge, this is the first study to apply safe screening techniques to worst-case robust models in supervised machine learning. Our approach safely identifies training samples whose uncertainty sets are guaranteed to lie entirely on either side of the margin hyperplane, thereby reducing the problem size and accelerating optimization. Owing to the nonstandard structure of R-SVMs, the proposed screening rules are derived from the Lagrangian duality rather than the Fenchel-Rockafellar duality commonly used in recent methods. Based on this analysis, we first establish an ideal screening rule, and then derive a practical rule by adapting GAP-based safe regions to the robust setting. Experiments demonstrate that the proposed method significantly reduces training time while preserving classification accuracy.

Integrating AI into the physical layer is a cornerstone of 6G networks. However, current data-driven approaches struggle to generalize across dynamic environments because they lack an intrinsic understanding of electromagnetic wave propagation. We introduce the Wireless World Model (WWM), a multi-modal foundation framework predicting the spatiotemporal evolution of wireless channels by internalizing the causal relationship between 3D geometry and signal dynamics. Pre-trained on a massive ray-traced multi-modal dataset, WWM overcomes the data authenticity gap, further validated under real-world measurement data. Using a joint-embedding predictive architecture with a multi-modal mixture-of-experts Transformer, WWM fuses channel state information, 3D point clouds, and user trajectories into a unified representation. Across the five key downstream tasks supported by WWM, it achieves remarkable performance in seen environments, unseen generalization scenarios, and real-world measurements, consistently outperforming SOTA uni-modal foundation models and task-specific models. This paves the way for physics-aware 6G intelligence that adapts to the physical world.

Generating long videos using pre-trained video diffusion models, which are typically trained on short clips, presents a significant challenge. Directly applying these models for long-video inference often leads to a notable degradation in visual quality. This paper identifies that this issue primarily stems from two out-of-distribution (O.O.D) problems: frame-level relative position O.O.D and context-length O.O.D. To address these challenges, we propose FreeLOC, a novel training-free, layer-adaptive framework that introduces two core techniques: Video-based Relative Position Re-encoding (VRPR) for frame-level relative position O.O.D, a multi-granularity strategy that hierarchically re-encodes temporal relative positions to align with the model's pre-trained distribution, and Tiered Sparse Attention (TSA) for context-length O.O.D, which preserves both local detail and long-range dependencies by structuring attention density across different temporal scales. Crucially, we introduce a layer-adaptive probing mechanism that identifies the sensitivity of each transformer layer to these O.O.D issues, allowing for the selective and efficient application of our methods. Extensive experiments demonstrate that our approach significantly outperforms existing training-free methods, achieving state-of-the-art results in both temporal consistency and visual quality. Code is available at https://github.com/Westlake-AGI-Lab/FreeLOC.

Conditional density estimation (CDE) is a fundamental task in machine learning that aims to model the full conditional law $\mathbb{P}(\mathbf{y} \mid \mathbf{x})$, beyond mere point prediction (e.g., mean, mode). A core challenge is free-form density estimation, capturing distributions that exhibit multimodality, asymmetry, or topological complexity without restrictive assumptions. However, prevailing methods typically estimate the probability density function (PDF) directly, which is mathematically ill-posed: differentiating the empirical distribution amplifies random fluctuations inherent in finite datasets, necessitating strong inductive biases that limit expressivity and fail when violated. We propose a CDF-first framework that circumvents this issue by estimating the cumulative distribution function (CDF), a stable and well-posed target, and then recovering the PDF via differentiation of the learned smooth CDF. Parameterizing the CDF with a Smooth Min-Max (SMM) network, our framework guarantees valid PDFs by construction, enables tractable approximate likelihood training, and preserves complex distributional shapes. For multivariate outputs, we use an autoregressive decomposition with SMM factors. Experiments demonstrate our approach outperforms state-of-the-art density estimators on a range of univariate and multivariate tasks.

Multimodal misinformation poses an escalating challenge that often evades traditional detectors, which are opaque black boxes and fragile against new manipulation tactics. We present Probabilistic Concept Graph Reasoning (PCGR), an interpretable and evolvable framework that reframes multimodal misinformation detection (MMD) as structured and concept-based reasoning. PCGR follows a build-then-infer paradigm, which first constructs a graph of human-understandable concept nodes, including novel high-level concepts automatically discovered and validated by multimodal large language models (MLLMs), and then applies hierarchical attention over this concept graph to infer claim veracity. This design produces interpretable reasoning chains linking evidence to conclusions. Experiments demonstrate that PCGR achieves state-of-the-art MMD accuracy and robustness to emerging manipulation types, outperforming prior methods in both coarse detection and fine-grained manipulation recognition.

Recent research points toward LLMs being manipulated through adversarial and seemingly benign inputs, resulting in harmful, biased, or policy-violating outputs. In this paper, we study an underexplored issue concerning harmful and toxic mathematical word problems. We show that math questions, particularly those framed as natural language narratives, can serve as a subtle medium for propagating biased, unethical, or psychologically harmful content, with heightened risks in educational settings involving children. To support a systematic study of this phenomenon, we introduce ToxicGSM, a dataset of 1.9k arithmetic problems in which harmful or sensitive context is embedded while preserving mathematically well-defined reasoning tasks. Using this dataset, we audit the behaviour of existing LLMs and analyse the trade-offs between safety enforcement and mathematical correctness. We further propose SafeMath -- a safety alignment technique that reduces harmful outputs while maintaining, and in some cases improving, mathematical reasoning performance. Our results highlight the importance of disentangling linguistic harm from math reasoning and demonstrate that effective safety alignment need not come at the cost of accuracy. We release the source code and dataset at https://github.com/Swagnick99/SafeMath/tree/main.

As AI assistants become integrated into safety engineering workflows for Physical AI systems, a critical question emerges: does AI assistance improve safety analysis quality, or introduce systematic blind spots that surface only through post-deployment incidents? This paper develops a formal framework for AI assistance in safety analysis. We first establish why safety engineering resists benchmark-driven evaluation: safety competence is irreducibly multidimensional, constrained by context-dependent correctness, inherent incompleteness, and legitimate expert disagreement. We formalize this through a five-dimensional competence framework capturing domain knowledge, standards expertise, operational experience, contextual understanding, and judgment. We introduce the competence shadow: the systematic narrowing of human reasoning induced by AI-generated safety analysis. The shadow is not what the AI presents, but what it prevents from being considered. We formalize four canonical human-AI collaboration structures and derive closed-form performance bounds, demonstrating that the competence shadow compounds multiplicatively to produce degradation far exceeding naive additive estimates. The central finding is that AI assistance in safety engineering is a collaboration design problem, not a software procurement decision. The same tool degrades or improves analysis quality depending entirely on how it is used. We derive non-degradation conditions for shadow-resistant workflows and call for a shift from tool qualification toward workflow qualification for trustworthy Physical AI.

Clinical practice guidelines (CPGs) play a pivotal role in ensuring evidence-based decision-making and improving patient outcomes. While Large Language Models (LLMs) are increasingly deployed in healthcare scenarios, it is unclear to which extend LLMs could identify and adhere to CPGs during conversations. To address this gap, we introduce CPGBench, an automated framework benchmarking the clinical guideline detection and adherence capabilities of LLMs in multi-turn conversations. We collect 3,418 CPG documents from 9 countries/regions and 2 international organizations published in the last decade spanning across 24 specialties. From these documents, we extract 32,155 clinical recommendations with corresponding publication institute, date, country, specialty, recommendation strength, evidence level, etc. One multi-turn conversation is generated for each recommendation accordingly to evaluate the detection and adherence capabilities of 8 leading LLMs. We find that the 71.1%-89.6% recommendations can be correctly detected, while only 3.6%-29.7% corresponding titles can be correctly referenced, revealing the gap between knowing the guideline contents and where they come from. The adherence rates range from 21.8% to 63.2% in different models, indicating a large gap between knowing the guidelines and being able to apply them. To confirm the validity of our automatic analysis, we further conduct a comprehensive human evaluation involving 56 clinicians from different specialties. To our knowledge, CPGBench is the first benchmark systematically revealing which clinical recommendations LLMs fail to detect or adhere to during conversations. Given that each clinical recommendation may affect a large population and that clinical applications are inherently safety critical, addressing these gaps is crucial for the safe and responsible deployment of LLMs in real world clinical practice.

Recent research on dialectal NLP has identified data scarcity as a primary limitation. To address this limitation, this paper presents a catalog of contemporary Basque dialectal data and resources, offering a systematic and comprehensive compilation of the dialectal data currently available in Basque. Two types of data sources have been distinguished: online data originally written in some dialect, and standard-to-dialect adapted data. The former includes all dialectal data that can be found online, such as news and radio sites, informal tweets, as well as online resources such as dictionaries, atlases, grammar rules, or videos. The latter consists of data that has been adapted from the standard variety to dialectal varieties, either manually or automatically. Regarding the manual adaptation, the test split of the XNLI Natural Language Inference dataset was manually adapted into three Basque dialects: Western, Central, and Navarrese-Lapurdian, yielding a high-quality parallel gold standard evaluation dataset. With respect to the automatic dialectal adaptation, the automatically adapted physical commonsense dataset (BasPhyCowest) underwent additional manual evaluation by native speakers to assess its quality and determine whether it could serve as a viable substitute for full manual adaptation (i.e., silver data creation).

Persona-driven large language models (LLMs) require consistent behavioral tendencies across interactions to simulate human-like personality traits, such as persistence or reliability. However, current LLMs often lack stable internal representations that anchor their responses over extended dialogues. This work explores whether LLMs can maintain "implicit consistency", defined as persistent adherence to an unstated goal in multi-turn interactions. We designed a 20-question-style riddle game paradigm where an LLM is tasked with secretly selecting a target and responding to users' guesses with "yes/no" answers. Through evaluations, we find that LLMs struggle to preserve latent consistency: their implicit "goals" shift across turns unless explicitly provided their selected target in context. These findings highlight critical limitations in the building of persona-driven LLMs and underscore the need for mechanisms that anchor implicit goals over time, which is a key to realistic personality modeling in interactive applications such as dialogue systems.

AI systems in healthcare research have shown potential to increase patient throughput and assist clinicians, yet progress is constrained by limited access to real patient data. To address this issue, we present a zero-shot, knowledge-guided framework for psychiatric tabular data in which large language models (LLMs) are steered via Retrieval-Augmented Generation using the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) and the International Classification of Diseases (ICD-10). We conducted experiments using different combinations of knowledge bases to generate privacy-preserving synthetic data. The resulting models were benchmarked against two state-of-the-art deep learning models for synthetic tabular data generation, namely CTGAN and TVAE, both of which rely on real data and therefore entail potential privacy risks. Evaluation was performed on six anxiety-related disorders: specific phobia, social anxiety disorder, agoraphobia, generalized anxiety disorder, separation anxiety disorder, and panic disorder. CTGAN typically achieves the best marginals and multivariate structure, while the knowledge-augmented LLM is competitive on pairwise structure and attains the lowest pairwise error in separation anxiety and social anxiety. An ablation study shows that clinical retrieval reliably improves univariate and pairwise fidelity over a no-retrieval LLM. Privacy analyses indicate that the real data-free LLM yields modest overlaps and a low average linkage risk comparable to CTGAN, whereas TVAE exhibits extensive duplication despite a low k-map score. Overall, grounding an LLM in clinical knowledge enables high-quality, privacy-preserving synthetic psychiatric data when real datasets are unavailable or cannot be shared.

Reinforcement learning (RL) has become essential for post-training large language models (LLMs) in reasoning tasks. While scaling rollouts can stabilize training and enhance performance, the computational overhead is a critical issue. In algorithms like GRPO, multiple rollouts per prompt incur prohibitive costs, as a large portion of prompts provide negligible gradients and are thus of low utility. To address this problem, we investigate how to select high-utility prompts before the rollout phase. Our experimental analysis reveals that sample utility is non-uniform and evolving: the strongest learning signals concentrate at the ``learning edge", the intersection of intermediate difficulty and high uncertainty, which shifts as training proceeds. Motivated by this, we propose HIVE (History-Informed and online-VErified prompt selection), a dual-stage framework for data-efficient RL. HIVE utilizes historical reward trajectories for coarse selection and employs prompt entropy as a real-time proxy to prune instances with stale utility. By evaluating HIVE across multiple math reasoning benchmarks and models, we show that HIVE yields significant rollout efficiency without compromising performance.

Context-aware machine translation (MT) leverages document-level information, yet it does not consistently outperform sentence-level MT, as contextual signals are unevenly beneficial across sentences. Existing training objectives do not explicitly model this variability, limiting a model's ability to adaptively exploit context. In this paper, we propose Cross-Preference Learning (CPL), a preference-based training framework that explicitly captures the complementary benefits of sentence-level and context-aware MT. CPL achieves this by integrating both intra- and cross-condition preferences into the preference optimization objective. The introduction of intra- and cross-condition preferences provides explicit supervision on when and how contextual information improves translation quality. We validate the proposed approach on several public context-aware MT tasks using multiple models, including Qwen3-4B, Qwen3-8B, and Llama-3-8B. Experimental results demonstrate consistent improvements in translation quality and robustness across both input conditions, achieved without any architectural modifications.

Text-to-motion generation holds significant potential for cross-linguistic applications, yet it is hindered by the lack of bilingual datasets and the poor cross-lingual semantic understanding of existing language models. To address these gaps, we introduce BiHumanML3D, the first bilingual text-to-motion benchmark, constructed via LLM-assisted annotation and rigorous manual correction. Furthermore, we propose a simple yet effective baseline, Bilingual Motion Diffusion (BiMD), featuring Cross-Lingual Alignment (CLA). CLA explicitly aligns semantic representations across languages, creating a robust conditional space that enables high-quality motion generation from bilingual inputs, including zero-shot code-switching scenarios. Extensive experiments demonstrate that BiMD with CLA achieves an FID of 0.045 vs. 0.169 and R@3 of 82.8\% vs. 80.8\%, significantly outperforms monolingual diffusion models and translation baselines on BiHumanML3D, underscoring the critical necessity and reliability of our dataset and the effectiveness of our alignment strategy for cross-lingual motion synthesis. The dataset and code are released at \href{https://wengwanjiang.github.io/BilingualT2M-page}{https://wengwanjiang.github.io/BilingualT2M-page}

Prompt attacks, including jailbreaks and prompt injections, pose a critical security risk to Large Language Model (LLM) systems. In production, guardrails must mitigate these attacks under strict low-latency constraints, resulting in a deployment gap in which lightweight classifiers and rule-based systems struggle to generalize under distribution shift, while high-capacity LLM-based judges remain too slow or costly for live enforcement. In this work, we examine whether lightweight, general-purpose LLMs can reliably serve as security judges under real-world production constraints. Through careful prompt and output design, lightweight LLMs are guided through a structured reasoning process involving explicit intent decomposition, safety-signal verification, harm assessment, and self-reflection. We evaluate our method on a curated dataset combining benign queries from real-world chatbots with adversarial prompts generated via automated red teaming (ART), covering diverse and evolving patterns. Our results show that general-purpose LLMs, such as gemini-2.0-flash-lite-001, can serve as effective low-latency judges for live guardrails. This configuration is currently deployed in production as a centralized guardrail service for public service chatbots in Singapore. We additionally evaluate a Mixture-of-Models (MoM) setting to assess whether aggregating multiple LLM judges improves prompt-attack detection performance relative to single-model judges, with only modest gains observed.

In complex environments, infrared object detection exhibits broad applicability and stability across diverse scenarios. However, infrared object detection is vulnerable to both common corruptions and adversarial examples, leading to potential security risks. To improve the robustness of infrared object detection, current methods mostly adopt a data-driven ideology, which only superficially drives the network to fit the training data without specifically considering the unique characteristics of infrared images, resulting in limited robustness. In this paper, we revisit infrared physical knowledge and find that relative thermal radiation relations between different classes can be regarded as a reliable knowledge source under the complex scenarios of adversarial examples and common corruptions. Thus, we theoretically model thermal radiation relations based on the rank order of gray values for different classes, and further quantify the stability of various inter-class thermal radiation relations. Based on the above theoretical framework, we propose Knowledge-Guided Adversarial Training (KGAT) for infrared object detection, in which infrared physical knowledge is embedded into the adversarial training process, and the predicted results are optimized to be consistent with the actual physical laws. Extensive experiments on three infrared datasets and six mainstream infrared object detection models demonstrate that KGAT effectively enhances both clean accuracy and robustness against adversarial attacks and common corruptions.

LLM performance is highly sensitive to prompt design, yet whether automatic prompt optimization can replace expert prompt engineering in linguistic tasks remains unexplored. We present the first systematic comparison of hand-crafted zero-shot expert prompts, base DSPy signatures, and GEPA-optimized DSPy signatures across translation, terminology insertion, and language quality assessment, evaluating five model configurations. Results are task-dependent. In terminology insertion, optimized and manual prompts produce mostly statistically indistinguishable quality. In translation, each approach wins on different models. In LQA, expert prompts achieve stronger error detection while optimization improves characterization. Across all tasks, GEPA elevates minimal DSPy signatures, and the majority of expert-optimized comparisons show no statistically significant difference. We note that the comparison is asymmetric: GEPA optimization searches programmatically over gold-standard splits, whereas expert prompts require in principle no labeled data, relying instead on domain expertise and iterative refinement.

Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of applications. However, their practical deployment is often hindered by issues such as outdated knowledge and the tendency to generate hallucinations. To address these limitations, Retrieval-Augmented Generation (RAG) systems have been introduced, enhancing LLMs with external, up-to-date knowledge sources. Despite their advantages, RAG systems remain vulnerable to adversarial attacks, with data poisoning emerging as a prominent threat. Existing poisoning-based attacks typically require prior knowledge of the user's specific queries, limiting their flexibility and real-world applicability. In this work, we propose PIDP-Attack, a novel compound attack that integrates prompt injection with database poisoning in RAG. By appending malicious characters to queries at inference time and injecting a limited number of poisoned passages into the retrieval database, our method can effectively manipulate LLM response to arbitrary query without prior knowledge of the user's actual query. Experimental evaluations across three benchmark datasets (Natural Questions, HotpotQA, MS-MARCO) and eight LLMs demonstrate that PIDP-Attack consistently outperforms the original PoisonedRAG. Specifically, our method improves attack success rates by 4% to 16% on open-domain QA tasks while maintaining high retrieval precision, proving that the compound attack strategy is both necessary and highly effective.

Equipping Large Language Model (LLM) agents with domain-specific skills is critical for tackling complex tasks. Yet, manual authoring creates a severe scalability bottleneck. Conversely, automated skill generation often yields fragile or fragmented results because it either relies on shallow parametric knowledge or sequentially overfits to non-generalizable trajectory-local lessons. To overcome this, we introduce Trace2Skill, a framework that mirrors how human experts author skills: by holistically analyzing broad execution experience before distilling it into a single, comprehensive guide. Instead of reacting sequentially to individual trajectories, Trace2Skill dispatches a parallel fleet of sub-agents to analyze a diverse pool of executions. It extracts trajectory-specific lessons and hierarchically consolidates them into a unified, conflict-free skill directory via inductive reasoning. Trace2Skill supports both deepening existing human-written skills and creating new ones from scratch. Experiments in challenging domains, such as spreadsheet, VisionQA and math reasoning, show that Trace2Skill significantly improves upon strong baselines, including Anthropic's official xlsx skills. Crucially, this trajectory-grounded evolution does not merely memorize task instances or model-specific quirks: evolved skills transfer across LLM scales and generalize to OOD settings. For example, skills evolved by Qwen3.5-35B on its own trajectories improved a Qwen3.5-122B agent by up to 57.65 absolute percentage points on WikiTableQuestions. Ultimately, our results demonstrate that complex agent experience can be packaged into highly transferable, declarative skills -- requiring no parameter updates, no external retrieval modules, and utilizing open-source models as small as 35B parameters.

Recent vision and multimodal foundation backbones, such as Transformer families and state-space models like Mamba, have achieved remarkable progress, enabling unified modeling across images, text, and beyond. Despite their empirical success, these architectures remain far from the computational principles of the human brain, often demanding enormous amounts of training data while offering limited interpretability. In this work, we propose the Vision Hopfield Memory Network (V-HMN), a brain-inspired foundation backbone that integrates hierarchical memory mechanisms with iterative refinement updates. Specifically, V-HMN incorporates local Hopfield modules that provide associative memory dynamics at the image patch level, global Hopfield modules that function as episodic memory for contextual modulation, and a predictive-coding-inspired refinement rule for iterative error correction. By organizing these memory-based modules hierarchically, V-HMN captures both local and global dynamics in a unified framework. Memory retrieval exposes the relationship between inputs and stored patterns, making decisions more interpretable, while the reuse of stored patterns improves data efficiency. This brain-inspired design therefore enhances interpretability and data efficiency beyond existing self-attention- or state-space-based approaches. We conducted extensive experiments on public computer vision benchmarks, and V-HMN achieved competitive results against widely adopted backbone architectures, while offering better interpretability, higher data efficiency, and stronger biological plausibility. These findings highlight the potential of V-HMN to serve as a next-generation vision foundation model, while also providing a generalizable blueprint for multimodal backbones in domains such as text and audio, thereby bridging brain-inspired computation with large-scale machine learning.

Multimodal large language models are promising for clinical visual question answering tasks, but scaling to 3D imaging is hindered by high computational costs. Prior methods often rely on 2D slices or fixed-length token compression, disrupting volumetric continuity and obscuring subtle findings. We present Photon, a framework that represents 3D medical volumes with token sequences of variable length. Photon introduces instruction-conditioned token scheduling and surrogate gradient propagation to adaptively reduce tokens during both training and inference, which lowers computational cost while mitigating the attention dilution caused by redundant tokens. It incorporates a custom backpropagation rule with gradient restoration to enable differentiable optimization despite discrete token drop. To stabilize token compression and ensure reliable use of visual evidence, Photon further applies regularization objectives that mitigate language-only bias and improve reliability. Experiments on diverse medical visual question answering tasks show that Photon achieves state-of-the-art accuracy while reducing resource usage and accelerating both training and inference.

Retrieval-Augmented Generation (RAG) systems face significant challenges in complex reasoning, multi-hop queries, and domain-specific QA. While existing GraphRAG frameworks have made progress in structural knowledge organization, they still have limitations in cross-industry adaptability, community report integrity, and retrieval performance. This paper proposes UniAI-GraphRAG, an enhanced framework built upon open-source GraphRAG. The framework introduces three core innovations: (1) Ontology-Guided Knowledge Extraction that uses predefined Schema to guide LLMs in accurately identifying domain-specific entities and relations; (2) Multi-Dimensional Community Clustering Strategy that improves community completeness through alignment completion, attribute-based clustering, and multi-hop relationship clustering; (3) Dual-Channel Graph Retrieval Fusion that balances QA accuracy and performance through hybrid graph and community retrieval. Evaluation results on MultiHopRAG benchmark show that UniAI-GraphRAG outperforms mainstream open source solutions (e.g.LightRAG) in comprehensive F1 scores, particularly in inference and temporal queries. The code is available at https://github.com/UnicomAI/wanwu/tree/main/rag/rag_open_source/rag_core/graph.

In speech evaluation, an Automatic Speech Recognition (ASR) model often computes time boundaries and phoneme posteriors for input features. However, limited data for ASR training hinders expansion of speech evaluation to low-resource languages. Open-source weakly-supervised models are capable of ASR over many languages, but they are frame-asynchronous and not phonemic, hindering feature extraction for speech evaluation. This paper proposes to overcome incompatibilities for feature extraction with weakly-supervised models, easing expansion of speech evaluation to low-resource languages. Phoneme posteriors are computed by mapping ASR hypotheses to a phoneme confusion network. Word instead of phoneme-level speaking rate and duration are used. Phoneme and frame-level features are combined using a cross-attention architecture, obviating phoneme time alignment. This performs comparably with standard frame-synchronous features on English speechocean762 and low-resource Tamil datasets.

Context: The rapid adoption of AI-assisted code generation tools, such as large language models (LLMs), is transforming software development practices. While these tools promise significant productivity gains, concerns regarding the quality, reliability, and security of AI-generated code are increasingly reported in both academia and industry. --Objective: This study aims to systematically synthesize existing empirical evidence on the factors influencing the quality of AI-generated source code and to analyze how these factors impact software quality outcomes across different evaluation contexts. --Method: We conducted a systematic literature review (SLR) following established guidelines, supported by an AI-assisted workflow with human oversight. A total of 24 primary studies were selected through a structured search and screening process across major digital libraries. Data were extracted and analyzed using qualitative, pattern-based evidence synthesis. --Results: The findings reveal that code quality in AI-assisted development is influenced by a combination of human factors, AI system characteristics, and human AI interaction dynamics. Key influencing factors include prompt design, task specification, and developer expertise. The results also show variability in quality outcomes such as correctness, security, maintainability, and complexity across studies, with both improvements and risks reported. --Conclusion: AI-assisted code generation represents a socio-technical shift in software engineering, where achieving high-quality outcomes depends on both technological and human factors. While promising, AI-generated code requires careful validation and integration into development workflows.

Direct preference optimization (DPO) is an effective technique to train language models to generate preferred over dispreferred responses. However, this binary "winner-takes-all" approach is suboptimal for vision-language models whose response quality is highly dependent on visual content. In particular, a response may still be faithful to the visual inputs even if it is less preferable than an alternative. The standard Bradley-Terry DPO formulation lacks this nuance, upweighting winning responses without sufficient regard for whether the "losing" response still maintains high visual fidelity. In this work, we investigate ranking optimization as an alternative that more precisely situates responses' faithfulness to visual inputs. We focus on video-text alignment using detailed video captions, proposing a method to generate challenging, totally ordered caption chains at scale through repeated caption degradation. Our results show ranking optimization outperforms binary DPO for long-form content generation and assessment, and importantly, we find that these approaches require finetuning of the vision encoder to be effective, challenging the view of DPO as purely a language-reweighting process.

Dataset distillation (DD) compresses a large training set into a small synthetic set, reducing storage and training cost, and has shown strong results on general benchmarks. Decoupled DD further improves efficiency by splitting the pipeline into pretraining, sample distillation, and soft-label generation. However, existing decoupled methods largely rely on coarse class-label supervision and optimize samples within each class in a nearly identical manner. On fine-grained datasets, this often yields distilled samples that (i) retain large intra-class variation with subtle inter-class differences and (ii) become overly similar within the same class, limiting localized discriminative cues and hurting recognition. To solve the above-mentioned problems, we propose FD$^{2}$, a dedicated framework for Fine-grained Dataset Distillation. FD$^{2}$ localizes discriminative regions and constructs fine-grained representations for distillation. During pretraining, counterfactual attention learning aggregates discriminative representations to update class prototypes. During distillation, a fine-grained characteristic constraint aligns each sample with its class prototype while repelling others, and a similarity constraint diversifies attention across same-class samples. Experiments on multiple fine-grained and general datasets show that FD$^{2}$ integrates seamlessly with decoupled DD and improves performance in most settings, indicating strong transferability.

Multimodal deepfakes can exhibit subtle visual artifacts and cross-modal inconsistencies, which remain challenging to detect, especially when detectors are trained primarily on curated synthetic forgeries. Such synthetic dependence can introduce dataset and generator bias, limiting scalability and robustness to unseen manipulations. We propose SAVe, a self-supervised audio-visual deepfake detection framework that learns entirely on authentic videos. SAVe generates on-the-fly, identity-preserving, region-aware self-blended pseudo-manipulations to emulate tampering artifacts, enabling the model to learn complementary visual cues across multiple facial granularities. To capture cross-modal evidence, SAVe also models lip-speech synchronization via an audio-visual alignment component that detects temporal misalignment patterns characteristic of audio-visual forgeries. Experiments on FakeAVCeleb and AV-LipSync-TIMIT demonstrate competitive in-domain performance and strong cross-dataset generalization, highlighting self-supervised learning as a scalable paradigm for multimodal deepfake detection.

In reinforcement learning, an agent interacts sequentially with an environment to maximize a reward, receiving only partial, probabilistic feedback. This creates a fundamental exploration-exploitation trade-off: the agent must explore to learn the hidden dynamics while exploiting this knowledge to maximize its target objective. While extensively studied classically, applying this framework to quantum systems requires dealing with hidden quantum states that evolve via unknown dynamics. We formalize this problem via a framework where the environment maintains a hidden quantum memory evolving via unknown quantum channels, and the agent intervenes sequentially using quantum instruments. For this setting, we adapt an optimistic maximum-likelihood estimation algorithm. We extend the analysis to continuous action spaces, allowing us to model general positive operator-valued measures (POVMs). By controlling the propagation of estimation errors through quantum channels and instruments, we prove that the cumulative regret of our strategy scales as $\widetilde{\mathcal{O}}(\sqrt{K})$ over $K$ episodes. Furthermore, via a reduction to the multi-armed quantum bandit problem, we establish information-theoretic lower bounds demonstrating that this sublinear scaling is strictly optimal up to polylogarithmic factors. As a physical application, we consider state-agnostic work extraction. When extracting free energy from a sequence of non-i.i.d. quantum states correlated by a hidden memory, any lack of knowledge about the source leads to thermodynamic dissipation. In our setting, the mathematical regret exactly quantifies this cumulative dissipation. Using our adaptive algorithm, the agent uses past energy outcomes to improve its extraction protocol on the fly, achieving sublinear cumulative dissipation, and, consequently, an asymptotically zero dissipation rate.

Rubric-based evaluation has become a prevailing paradigm for evaluating instruction following in large language models (LLMs). Despite its widespread use, the reliability of these rubric-level evaluations remains unclear, calling for meta-evaluation. However, prior meta-evaluation efforts largely focus on the response level, failing to assess the fine-grained judgment accuracy that rubric-based evaluation relies on. To bridge this gap, we introduce RubricEval. Our benchmark features: (1) the first rubric-level meta-evaluation benchmark for instruction following, (2) diverse instructions and responses spanning multiple categories and model sources, and (3) a substantial set of 3,486 quality-controlled instances, along with Easy/Hard subsets that better differentiates judge performance. Our experiments reveal that rubric-level judging remains far from solved: even GPT-4o, a widely adopted judge in instruction-following benchmarks, achieves only 55.97% on Hard subset. Considering evaluation paradigm, rubric-level evaluation outperforms checklist-level, explicit reasoning improves accuracy, and both together reduce inter-judge variance. Through our established rubric taxonomy, we further identify common failure modes and offer actionable insights for reliable instruction-following evaluation.

Robust principal component analysis (RPCA) seeks a low-rank component and a sparse component from their summation. Yet, in many applications of interest, the sparse foreground actually replaces, or occludes, elements from the low-rank background. To address this mismatch, a new framework is proposed in which the sparse component is identified indirectly through determining its support. This approach, called robust principal component completion (RPCC), is solved via variational Bayesian inference applied to a fully probabilistic Bayesian sparse tensor factorization. Convergence to a hard classifier for the support is shown, thereby eliminating the post-hoc thresholding required of most prior RPCA-driven approaches. Experimental results reveal that the proposed approach delivers near-optimal estimates on synthetic data as well as robust foreground-extraction and anomaly-detection performance on real color video and hyperspectral datasets, respectively. Source implementation and Appendices are available at https://github.com/WongYinJ/BCP-RPCC.

Multi-Behavior Recommendation (MBR) leverages multiple user interaction types (e.g., views, clicks, purchases) to enrich preference modeling and alleviate data sparsity issues in traditional single-behavior approaches. However, existing MBR methods face fundamental challenges: they lack principled frameworks to model complex confounding effects from user behavioral habits and item multi-behavior distributions, struggle with effective aggregation of heterogeneous auxiliary behaviors, and fail to align behavioral representations across semantic gaps while accounting for bias distortions. To address these limitations, we propose MCLMR, a novel model-agnostic causal learning framework that can be seamlessly integrated into various MBR architectures. MCLMR first constructs a causal graph to model confounding effects and performs interventions for unbiased preference estimation. Under this causal framework, it employs an Adaptive Aggregation module based on Mixture-of-Experts to dynamically fuse auxiliary behavior information and a Bias-aware Contrastive Learning module to align cross-behavior representations in a bias-aware manner. Extensive experiments on three real-world datasets demonstrate that MCLMR achieves significant performance improvements across various baseline models, validating its effectiveness and generality. All data and code will be made publicly available. For anonymous review, our code is available at the following the link: https://github.com/gitrxh/MCLMR.

Few-Shot Class-Incremental Learning (FSCIL) can be particularly susceptible to acquisition contexts with only a few labeled samples. A typical scenario is tactile sensing, where the acquisition context ({\it e.g.}, diverse devices, contact state, and interaction settings) degrades performance due to a lack of standardization. In this paper, we propose Context-as-Transform FSCIL (CaT-FSCIL) to tackle the above problem. We decompose the acquisition context into a structured low-dimensional component and a high-dimensional residual component. The former can be easily affected by tactile interaction features, which are modeled as an approximately invertible Context-as-Transform family and handled via inverse-transform canonicalization optimized with a pseudo-context consistency loss. The latter mainly arises from platform and device differences, which can be mitigated with an Uncertainty-Conditioned Prototype Calibration (UCPC) that calibrates biased prototypes and decision boundaries based on context uncertainty. Comprehensive experiments on the standard benchmarks HapTex and LMT108 have demonstrated the superiority of the proposed CaT-FSCIL.

Standard evaluation of LLM confidence relies on calibration metrics (ECE, Brier score) that conflate two distinct capacities: how much a model knows (Type-1 sensitivity) and how well it knows what it knows (Type-2 metacognitive sensitivity). We introduce an evaluation framework based on Type-2 Signal Detection Theory that decomposes these capacities using meta-d' and the metacognitive efficiency ratio M-ratio. Applied to four LLMs (Llama-3-8B-Instruct, Mistral-7B-Instruct-v0.3, Llama-3-8B-Base, Gemma-2-9B-Instruct) across 224,000 factual QA trials, we find: (1) metacognitive efficiency varies substantially across models even when Type-1 sensitivity is similar -- Mistral achieves the highest d' but the lowest M-ratio; (2) metacognitive efficiency is domain-specific, with different models showing different weakest domains, invisible to aggregate metrics; (3) temperature manipulation shifts Type-2 criterion while meta-d' remains stable for two of four models, dissociating confidence policy from metacognitive capacity; (4) AUROC_2 and M-ratio produce fully inverted model rankings, demonstrating these metrics answer fundamentally different evaluation questions. The meta-d' framework reveals which models "know what they don't know" versus which merely appear well-calibrated due to criterion placement -- a distinction with direct implications for model selection, deployment, and human-AI collaboration. Pre-registered analysis; code and data publicly available.

Recent advances have shown the effectiveness of self-evolving LLM agents on tasks such as program repair and scientific discovery. In this paradigm, a planner LLM synthesizes an agent program that invokes parametric models, including LLMs, which are then tuned per task to improve performance. However, existing self-evolving agent frameworks provide no formal guarantees of safety or correctness. Because such programs are often executed autonomously on unseen inputs, this lack of guarantees raises reliability and security concerns. We formulate agentic code generation as a constrained learning problem, combining hard formal specifications with soft objectives capturing task utility. We introduce Formally Guarded Generative Models (FGGM), which allow the planner LLM to specify a formal output contract for each generative model call using first-order logic. Each FGGM call wraps the underlying model in a rejection sampler with a verified fallback, ensuring every returned output satisfies the contract for any input and parameter setting. Building on FGGM, we present SEVerA (Self-Evolving Verified Agents), a three-stage framework: Search synthesizes candidate parametric programs containing FGGM calls; Verification proves correctness with respect to hard constraints for all parameter values, reducing the problem to unconstrained learning; and Learning applies scalable gradient-based optimization, including GRPO-style fine-tuning, to improve the soft objective while preserving correctness. We evaluate SEVerA on Dafny program verification, symbolic math synthesis, and policy-compliant agentic tool use ($τ^2$-bench). Across tasks, SEVerA achieves zero constraint violations while improving performance over unconstrained and SOTA baselines, showing that formal behavioral constraints not only guarantee correctness but also steer synthesis toward higher-quality agents.

Data augmentation is a key technique for improving the robustness of image classification models. However, many recent approaches rely on diffusion-based synthesis or complex feature mixing strategies, which introduce substantial computational overhead or require external datasets. In this work, we explore a different direction: procedural augmentation based on analytic interference patterns. Unlike conventional augmentation methods that rely on stochastic noise, feature mixing, or generative models, our approach exploits Moire interference to generate structured perturbations spanning a wide range of spatial frequencies. We propose a lightweight augmentation method that procedurally generates Moire textures on-the-fly using a closed-form mathematical formulation. The patterns are synthesized directly in memory with negligible computational cost (0.0026 seconds per image), mixed with training images during training, and immediately discarded, enabling a storage-free augmentation pipeline without external data. Extensive experiments with Vision Transformers demonstrate that the proposed method consistently improves robustness across multiple benchmarks, including ImageNet-C, ImageNet-R, and adversarial benchmarks, outperforming standard augmentation baselines and existing external-data-free augmentation approaches. These results suggest that analytic interference patterns provide a practical and efficient alternative to data-driven generative augmentation methods.

Large Language Models (LLMs) have shown remarkable capabilities for complex tasks, yet adaptation in medical domain, specifically mental health, poses specific challenges. Mental health is a rising concern globally with LLMs having large potential to help address the same. We highlight three primary challenges for LLMs in mental health - lack of high quality interpretable and knowledge grounded training data; training paradigms restricted to core capabilities, and evaluation of multi turn dialogue settings. Addressing it, we present oMind framework which includes training and aligning LLM agents for diverse capabilities including conversations; high quality ~164k multi-task SFT dataset, as a result of our generation pipeline based on Structured Knowledge retrieval, LLM based pruning, and review actions. We also introduce oMind-Chat - a novel multi turn benchmark dataset with expert annotated turn level and conversation level rubrics. Our diverse experiments on both core capabilities and conversations shows oMind LLMs consistently outperform baselines. oMind-LLM also shows significantly better reasoning with up to 80% win rate.

Modern multimodal systems deployed in industrial and safety-critical environments must remain reliable under partial sensor failures, signal degradation, or cross-modal inconsistencies. This work introduces a mathematically grounded framework for fault-tolerant multimodal representation learning that unifies self-supervised anomaly detection and error correction within a single architecture. Building upon a theoretical analysis of perturbation propagation, we derive Lipschitz- and Jacobian-based criteria that determine whether a neural operator amplifies or attenuates localized faults. Guided by this theory, we propose a two-stage self-supervised training scheme: pre-training a multimodal convolutional autoencoder on clean data to preserve localized anomaly signals in the latent space, and expanding it with a learnable compute block composed of dense layers for correction and contrastive objectives for anomaly identification. Furthermore, we introduce layer-specific Lipschitz modulation and gradient clipping as principled mechanisms to control sensitivity across detection and correction modules. Experimental results on multimodal fault datasets demonstrate that the proposed approach improves both anomaly detection accuracy and reconstruction under sensor corruption. Overall, this framework bridges the gap between analytical robustness guarantees and practical fault-tolerant multimodal learning.

Existing multi-agent frameworks allow each agent to simultaneously plan, execute, and evaluate its own actions -- a structural deficiency we term the "Logic Monopoly." Empirical evidence quantifies the resulting "Reliability Gap": 84.30% average attack success rates across ten deployment scenarios, 31.4% emergent deceptive behavior without explicit reward signals, and cascading failure modes rooted in six structural bottlenecks. The remedy is not better alignment of individual models but a social contract for agents: institutional infrastructure that enforces a constitutional Separation of Power. This paper introduces the Agent Enterprise for Enterprise (AE4E) paradigm -- agents as autonomous, legally identifiable business entities within a functionalist social system -- with a contract-centric SoP model trifurcating authority into Legislation, Execution, and Adjudication branches. The paradigm is operationalized through the NetX Enterprise Framework (NEF): governance hubs, TEE-backed compute enclaves, privacy-preserving data bridges, and an Agent-Native blockchain substrate. The Agent Enterprise Economy scales across four deployment tiers from private enclaves to a global Web of Services. The Agentic Social Layer, grounded in Parsons' AGIL framework, provides institutional infrastructure via sixty-plus named Institutional AE4Es. 143 pages, 173 references, eight specialized smart contracts.

We present a framework in which a large language model (LLM) acts as an online adaptive controller for SIMP topology optimization, replacing conventional fixed-schedule continuation with real-time, state-conditioned parameter decisions. At every $k$-th iteration, the LLM receives a structured observation$-$current compliance, grayness index, stagnation counter, checkerboard measure, volume fraction, and budget consumption$-$and outputs numerical values for the penalization exponent $p$, projection sharpness $β$, filter radius $r_{\min}$, and move limit $δ$ via a Direct Numeric Control interface. A hard grayness gate prevents premature binarization, and a meta-optimization loop uses a second LLM pass to tune the agent's call frequency and gate threshold across runs. We benchmark the agent against four baselines$-$fixed (no-continuation), standard three-field continuation, an expert heuristic, and a schedule-only ablation$-$on three 2-D problems (cantilever, MBB beam, L-bracket) at $120\!\times\!60$ resolution and two 3-D problems (cantilever, MBB beam) at $40\!\times\!20\!\times\!10$ resolution, all run for 300 iterations. A standardized 40-iteration sharpening tail is applied from the best valid snapshot so that compliance differences reflect only the exploration phase. The LLM agent achieves the lowest final compliance on every benchmark: $-5.7\%$ to $-18.1\%$ relative to the fixed baseline, with all solutions fully binary. The schedule-only ablation underperforms the fixed baseline on two of three problems, confirming that the LLM's real-time intervention$-$not the schedule geometry$-$drives the gain. Code and reproduction scripts will be released upon publication.

Large Language Model based agents increasingly operate in high stakes, multi turn settings where factual grounding is critical, yet their memory systems typically rely on flat key value stores or plain vector retrieval with no mechanism to track the provenance or trustworthiness of stored knowledge. We present ElephantBroker, an open source cognitive runtime that unifies a Neo4j knowledge graph with a Qdrant vector store through the Cognee SDK to provide durable, verifiable agent memory. The system implements a complete cognitive loop (store, retrieve, score, compose, protect, learn) comprising a hybrid five source retrieval pipeline, an eleven dimension competitive scoring engine for budget constrained context assembly, a four state evidence verification model, a five stage context lifecycle with goal aware assembly and continuous compaction, a six layer cheap first guard pipeline for safety enforcement, an AI firewall providing enforceable tool call interception and multi tier safety scanning, a nine stage consolidation engine that strengthens useful patterns while decaying noise, and a numeric authority model governing multi organization identity with hierarchical access control. Architectural validation through a comprehensive test suite of over 2,200 tests spanning unit, integration, and end to end levels confirms subsystem correctness. The modular design supports three deployment tiers, five profile presets with inheritance, multi gateway isolation, and a management dashboard for human oversight, enabling configurations from lightweight memory only agents to full cognitive runtimes with enterprise grade safety and auditability.

Machine learning models can achieve high predictive accuracy in hydrological applications but often lack physical interpretability. The Mass-Conserving Perceptron (MCP) provides a physics-aware artificial intelligence (AI) framework that enforces conservation principles while allowing hydrological process relationships to be learned from data. In this study, we investigate how progressively embedding physically meaningful representations of hydrological processes within a single MCP storage unit improves predictive skill and interpretability in rainfall-runoff modeling. Starting from a minimal MCP formulation, we sequentially introduce bounded soil storage, state-dependent conductivity, variable porosity, infiltration capacity, surface ponding, vertical drainage, and nonlinear water-table dynamics. The resulting hierarchy of process-aware MCP models is evaluated across 15 catchments spanning five hydroclimatic regions of the continental United States using daily streamflow prediction as the target. Results show that progressively augmenting the internal physical structure of the MCP unit generally improves predictive performance. The influence of these process representations is strongly hydroclimate dependent: vertical drainage substantially improves model skill in arid and snow-dominated basins but reduces performance in rainfall-dominated regions, while surface ponding has comparatively small effects. The best-performing MCP configurations approach the predictive skill of a Long Short-Term Memory benchmark while maintaining explicit physical interpretability. These results demonstrate that embedding hydrological process constraints within AI architectures provides a promising pathway toward interpretable and process-aware rainfall-runoff modeling.

Most vision-language systems are static observers: they describe pixels, do not act, and cannot safely improve under shift. This passivity limits generalizable, physically grounded visual intelligence. Learning through action, not static description, is essential beyond curated data. We present Pixelis, a pixel-space agent that operates directly on images and videos via a compact set of executable operations (zoom/crop, segment, track, OCR, temporal localization) and learns from its consequences. Pixelis trains in three phases: (1) Supervised Fine-Tuning learns a pixel-tool grammar from Chain-of-Thought-Action traces with a masked imitation loss that upweights operation/argument tokens and auxiliary heads to stabilize pixel-grounded arguments; (2) Curiosity-Coherence Reward Fine-Tuning optimizes a dual-drive objective marrying prediction-error curiosity with adjacent-step coherence and a mild efficiency prior under a KL anchor, yielding short, valid, structured toolchains; (3) Pixel Test-Time RL performs label-free adaptation by retrieving neighbors, voting over complete trajectories rather than answers, and updating toward short, high-fidelity exemplars while constraining drift with a KL-to-EMA safety control. Across six public image and video benchmarks, Pixelis yields consistent improvements: the average relative gain is +4.08% over the same 8B baseline (peaking at +6.03% on VSI-Bench), computed as (ours-baseline)/baseline, while producing shorter, auditable toolchains and maintaining in-corridor KL during test-time learning. Acting within pixels, rather than abstract tokens, grounds multimodal perception in the physical world, linking visual reasoning with actionable outcomes, and enables embodied adaptation without external feedback.

Out-of-Distribution (OOD) generalization has become a primary metric for evaluating image analysis systems. Since deep learning models tend to capture domain-specific context, they often develop shortcut dependencies on these non-causal features, leading to inconsistent performance across different data sources. Current techniques, such as invariance learning, attempt to mitigate this. However, they struggle to isolate highly mixed features within deep latent spaces. This limitation prevents them from fully resolving the shortcut learning problem.In this paper, we propose Hierarchical Causal Dropout (HCD), a method that uses channel-level causal masks to enforce feature sparsity. This approach allows the model to separate causal features from spurious ones, effectively performing a causal intervention at the representation level. The training is guided by a Matrix-based Mutual Information (MMI) objective to minimize the mutual information between latent features and domain labels, while simultaneously maximizing the information shared with class labels.To ensure stability, we incorporate a StyleMix-driven VICReg module, which prevents the masks from accidentally filtering out essential causal data. Experimental results on OOD benchmarks show that HCD performs better than existing top-tier methods.

Sparse autoencoders (SAEs) improve interpretability in multimodal models, but it remains unclear whether SAE features form modular, composable units for reasoning-an assumption underlying many intervention-based steering methods. We test this modularity hypothesis and find it often fails: intervening on a task-selective feature set can modestly improve reasoning accuracy, while intervening on the union of two such sets reliably induces output drift (large unintended changes in predictions) and degrades accuracy, even under norm-matched perturbations. This non modular circuit interference is consistent with shared internal pathways where feature unions amplify activation shifts. We develop a reproducible causal pipeline to localize and test these sparse visual thought circuits in Qwen3-VL-8B. On a controlled synthetic benchmark with seven task types and three difficulty levels, linear probes identify a mid decoder locus for task type information. We train SAEs at this layer, construct task-selective sets via an explicit rule, and perform inference time scaling and ablation while quantifying accuracy and drift. Our findings-validated with bootstrapped subsamples and permutation controls, and replicated across multiple VLM families and five diverse datasets clarify the boundaries of SAE feature composability and provide a rigorous diagnostic framework for more reliable VLM control.

Agriculture is increasingly challenged by climate change, soil degradation, and resource depletion, and hence requires advanced data-driven crop classification and recommendation solutions. This work presents an explainable ensemble learning paradigm that fuses optimized feature pyramids, deep networks, self-attention mechanisms, and residual networks for bolstering crop suitability predictions based on soil characteristics (e.g., pH, nitrogen, potassium) and climatic conditions (e.g., temperature, rainfall). With a dataset comprising 3,867 instances and 29 features from the Ethiopian Agricultural Transformation Agency and NASA, the paradigm leverages preprocessing methods such as label encoding, outlier removal using IQR, normalization through StandardScaler, and SMOTE for balancing classes. A range of machine learning models such as Logistic Regression, K-Nearest Neighbors, Support Vector Machines, Decision Trees, Random Forest, Gradient Boosting, and a new Relative Error Support Vector Machine are compared, with hyperparameter tuning through Grid Search and cross-validation. The suggested "Final Ensemble" meta-ensemble design outperforms with 98.80% accuracy, precision, recall, and F1-score, compared to individual models such as K-Nearest Neighbors (95.56% accuracy). Explainable AI methods, such as SHAP and permutation importance, offer actionable insights, highlighting critical features such as soil pH, nitrogen, and zinc. The paradigm addresses the gap between intricate ML models and actionable agricultural decision-making, fostering sustainability and trust in AI-powered recommendations

Traffic digital twins, which inform policymakers of effective interventions based on large-scale, high-fidelity computational models calibrated to real-world traffic, hold promise for addressing societal challenges in our rapidly urbanizing world. However, conventional fine-grained traffic simulations are non-differentiable and typically rely on inefficient gradient-free optimization, making calibration for real-world applications computationally infeasible. Here we present a differentiable agent-based traffic simulator that enables ultra-fast model calibration, traffic nowcasting, and control on large-scale networks. We develop several differentiable computing techniques for simulating individual vehicle movements, including stochastic decision-making and inter-agent interactions, while ensuring that entire simulation trajectories remain end-to-end differentiable for efficient gradient-based optimization. On the large-scale Chicago road network, with over 10,000 calibration parameters, our model simulates more than one million vehicles at 173 times real-time speed. This ultra-fast simulation, together with efficient gradient-based optimization, enables us to complete model calibration using the previous 30 minutes of traffic data in 455 s, provide a one-hour-ahead traffic nowcast in 21 s, and solve the resulting traffic control problem in 728 s. This yields a full calibration--nowcast--control loop in under 20 minutes, leaving about 40 minutes of lead time for implementing interventions. Our work thus provides a practical computational basis for realizing traffic digital twins.

Recent agentic systems demonstrate that large language models can generate scientific visualizations from natural language. However, reliability remains a major limitation: systems may execute invalid operations, introduce subtle but consequential errors, or fail to request missing information when inputs are underspecified. These issues are amplified in real-world workflows, which often exceed the complexity of standard benchmarks. Ensuring reliability in autonomous visualization pipelines therefore remains an open challenge. We present TopoPilot, a reliable and extensible agentic framework for automating complex scientific visualization workflows. TopoPilot incorporates systematic guardrails and verification mechanisms to ensure reliable operation. While we focus on topological data analysis and visualization as a primary use case, the framework is designed to generalize across visualization domains. TopoPilot adopts a reliability-centered two-agent architecture. An orchestrator agent translates user prompts into workflows composed of atomic backend actions, while a verifier agent evaluates these workflows prior to execution, enforcing structural validity and semantic consistency. This separation of interpretation and verification reduces code-generation errors and enforces correctness guarantees. A modular architecture further improves robustness by isolating components and enabling seamless integration of new descriptors and domain-specific workflows without modifying the core system. To systematically address reliability, we introduce a taxonomy of failure modes and implement targeted safeguards for each class. In evaluations simulating 1,000 multi-turn conversations across 100 prompts, including adversarial and infeasible requests, TopoPilot achieves a success rate exceeding 99%, compared to under 50% for baselines without comprehensive guardrails and checks.

Linearized string representations serve as the foundation of scalable autoregressive molecular generation; however, they introduce a fundamental modality mismatch where a single molecular graph maps to multiple distinct sequences. This ambiguity leads to \textit{trajectory divergence}, where the latent representations of structurally equivalent partial graphs drift apart due to differences in linearization history. To resolve this without abandoning the efficient string formulation, we propose Structure-Invariant Generative Molecular Alignment (SIGMA). Rather than altering the linear representation, SIGMA enables the model to strictly recognize geometric symmetries via a token-level contrastive objective, which explicitly aligns the latent states of prefixes that share identical suffixes. Furthermore, we introduce Isomorphic Beam Search (IsoBeam) to eliminate isomorphic redundancy during inference by dynamically pruning equivalent paths. Empirical evaluations on standard benchmarks demonstrate that SIGMA bridges the gap between sequence scalability and graph fidelity, yielding superior sample efficiency and structural diversity in multi-parameter optimization compared to strong baselines.

System prompt configuration can make the difference between near-total phishing blindness and near-perfect detection in LLM email agents. We present PhishNChips, a study of 11 models under 10 prompt strategies, showing that prompt-model interaction is a first-order security variable: a single model's phishing bypass rate ranges from under 1% to 97% depending on how it is configured, while the false-positive cost of the same prompt varies sharply across models. We then show that optimizing prompts around highly predictive signals can improve benchmark performance, reaching up to 93.7% recall at 3.8% false positive rate, but also creates a brittle attack surface. In particular, domain-matching strategies perform well when legitimate emails mostly have matched sender and URL domains, yet degrade sharply when attackers invert that signal by registering matching infrastructure. Response-trace analysis shows that 98% of successful bypasses reason in ways consistent with the inverted signal: the models are following the instruction, but the instruction's core assumption has become false. A counter-intuitive corollary follows: making prompts more specific can degrade already-capable models by replacing broader multi-signal reasoning with exploitable single-signal dependence. We characterize the resulting tension between detection, usability, and adversarial robustness as a navigable tradeoff, introduce Safetility, a deployability-aware metric that penalizes false positives, and argue that closing the adversarial gap likely requires tool augmentation with external ground truth.

Large language models (LLMs) tend to verbalize confidence scores that are largely detached from their actual accuracy, yet the geometric relationship governing this behavior remain poorly understood. In this work, we present a mechanistic interpretability analysis of verbalized confidence, using linear probes and contrastive activation addition (CAA) steering to show that calibration and verbalized confidence signals are encoded linearly but are orthogonal to one another -- a finding consistent across three open-weight models and four datasets. Interestingly, when models are prompted to simultaneously reason through a problem and verbalize a confidence score, the reasoning process disrupts the verbalized confidence direction, exacerbating miscalibration. We term this the "Reasoning Contamination Effect." Leveraging this insight, we introduce a two-stage adaptive steering pipeline that reads the model's internal accuracy estimate and steers verbalized output to match it, substantially improving calibration alignment across all evaluated models.

This study presents a computational analysis of the Slovene historical newspapers \textit{Slovenec} and \textit{Slovenski narod} from the sPeriodika corpus, combining topic modelling, large language model (LLM)-based aspect-level sentiment analysis, entity-graph visualisation, and qualitative discourse analysis to examine how collective identities, political orientations, and national belonging were represented in public discourse at the turn of the twentieth century. Using BERTopic, we identify major thematic patterns and show both shared concerns and clear ideological differences between the two newspapers, reflecting their conservative-Catholic and liberal-progressive orientations. We further evaluate four instruction-following LLMs for targeted sentiment classification in OCR-degraded historical Slovene and select the Slovene-adapted GaMS3-12B-Instruct model as the most suitable for large-scale application, while also documenting important limitations, particularly its stronger performance on neutral sentiment than on positive or negative sentiment. Applied at dataset scale, the model reveals meaningful variation in the portrayal of collective identities, with some groups appearing predominantly in neutral descriptive contexts and others more often in evaluative or conflict-related discourse. We then create NER graphs to explore the relationships between collective identities and places. We apply a mixed methods approach to analyse the named entity graphs, combining quantitative network analysis with critical discourse analysis. The investigation focuses on the emergence and development of intertwined historical political and socionomic identities. Overall, the study demonstrates the value of combining scalable computational methods with critical interpretation to support digital humanities research on noisy historical newspaper data.

Precipitation forecasting remains a persistent challenge in tropical regions like Vietnam, where complex topography and convective instability often limit the accuracy of Numerical Weather Prediction (NWP) models. While data-driven post-processing is widely used to mitigate these biases, most existing frameworks rely on point-wise objective functions, which suffer from the ``double penalty'' effect under minor temporal misalignments. In this work, we propose the Matrix Profile-guided Mixture of Experts (MP-MoE), a framework that integrates conventional intensity loss with a structural-aware Matrix Profile objective. By leveraging subsequence-level similarity rather than point-wise errors, the proposed loss facilitates more reliable expert selection and mitigates excessive penalization caused by phase shifts. We evaluate MP-MoE on rainfall datasets from two major river basins in Vietnam across multiple horizons, including 1-hour intensity and accumulated rainfall over 12, 24, and 48 hours. Experimental results demonstrate that MP-MoE outperforms raw NWP and baseline learning methods in terms of Mean Critical Success Index (CSI-M) for heavy rainfall events, while significantly reducing Dynamic Time Warping (DTW) values. These findings highlight the framework's efficacy in capturing peak rainfall intensities and preserving the morphological integrity of storm events.

We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, including chemistry, materials, life sciences, and earth sciences. Achieving this massive scale is made possible by the robust infrastructure support of XTuner and LMDeploy, which facilitates highly efficient Reinforcement Learning (RL) training at the 1-trillion parameter level while ensuring strict precision consistency between training and inference. By seamlessly integrating these advancements, Intern-S1-Pro further fortifies the fusion of general and specialized intelligence, working as a Specializable Generalist, demonstrating its position in the top tier of open-source models for general capabilities, while outperforming proprietary models in the depth of specialized scientific tasks.

Compressed vision-language models (VLMs) are widely used to reduce memory and compute costs, making them a suitable choice for real-world deployment. However, compressing these models raises concerns about whether internal computations and safety behaviors are preserved. In this work, we use causal circuit analysis and crosscoder-based feature comparisons to examine how pruning and quantization fundamentally change the internals across representative VLMs. We observe that pruning generally keeps circuit structure intact but rotates and attenuates internal features, while quantization modifies the circuits at a higher level yet leaves the surviving features better aligned. Leveraging this insight, we also introduce VLMSafe-420, a novel benchmark that pairs harmful inputs with matched benign counterfactuals across various safety categories. Our findings show that pruning causes a sharp drop in genuine refusal behavior, suggesting that the choice of compression has safety implications.

Foundation models excel in stable environments, yet often fail where reliability matters most: medicine, finance, and policy. This Fidelity Paradox is not just a data problem; it is structural. In domains where rules change over time, extra model capacity amplifies noise rather than capturing signal. We introduce Epistemic Compression: the principle that robustness emerges from matching model complexity to the shelf life of the data, not from scaling parameters. Unlike classical regularization, which penalizes weights post hoc, Epistemic Compression enforces parsimony through architecture: the model structure itself is designed to reduce overfitting by making it architecturally costly to represent variance that exceeds the evidence in the data. We operationalize this with a Regime Index that separates Shifting Regime (unstable, data-poor; simplicity wins) from Stable Regime (invariant, data-rich; complexity viable). In an exploratory synthesis of 15 high-stakes domains, this index was concordant with the empirically superior modeling strategy in 86.7% of cases (13/15). High-stakes AI demands a shift from scaling for its own sake to principled parsimony.

In psychological support and emotional companionship scenarios, the core limitation of large language models (LLMs) lies not merely in response quality, but in their reliance on local next-token prediction, which prevents them from maintaining the temporal continuity, stage awareness, and user consent boundaries required for multi-turn intervention. This stateless characteristic makes systems prone to premature advancement, stage misalignment, and boundary violations in continuous dialogue. To address this problem, we argue that the key challenge in process-oriented emotional support is not simply generating natural language, but constructing a sustainably updatable external situational structure for the model. We therefore propose LEKIA 2.0, a situated LLM architecture that separates the cognitive layer from the executive layer, thereby decoupling situational modeling from intervention execution. This design enables the system to maintain stable representations of the user's situation and consent boundaries throughout ongoing interaction. To evaluate this process-control capability, we further introduce a Static-to-Dynamic online evaluation protocol for multi-turn interaction. LEKIA achieved an average absolute improvement of approximately 31% over prompt-only baselines in deep intervention loop completion. The results suggest that an external situational structure is a key enabling condition for building stable, controllable, and situated emotional support systems.

We consider the problem of Online Convex Optimization (OCO) with two-point bandit feedback in an adversarial environment. In this setting, a player attempts to minimize a sequence of adversarially generated convex loss functions, while only observing the value of each function at two points. While it is well-known that two-point feedback allows for gradient estimation, achieving tight high-probability regret bounds for strongly convex functions still remained open as highlighted by \citet{agarwal2010optimal}. The primary challenge lies in the heavy-tailed nature of bandit gradient estimators, which makes standard concentration analysis difficult. In this paper, we resolve this open challenge by providing the first high-probability regret bound of $O(d(\log T + \log(1/δ))/μ)$ for $μ$-strongly convex losses. Our result is minimax optimal with respect to both the time horizon $T$ and the dimension $d$.

Autoregressive neural PDE simulators predict the evolution of physical fields one step at a time from a finite history, but low-cost context-window selection for such simulators remains an unformalized problem. Existing approaches to context-window selection in time-series forecasting include exhaustive validation, direct low-cost search, and system-theoretic memory estimation, but they are either expensive, brittle, or not directly aligned with downstream rollout performance. We formalize explicit context-window selection for fixed-window autoregressive neural PDE simulators as an independent low-cost algorithmic problem, and propose \textbf{System-Anchored Knee Estimation (SAKE)}, a two-stage method that first identifies a small structured candidate set from physically interpretable system anchors and then performs knee-aware downstream selection within it. Across all eight PDEBench families evaluated under the shared \(L\in\{1,\dots,16\}\) protocol, SAKE is the strongest overall matched-budget low-cost selector among the evaluated methods, achieving 67.8\% Exact, 91.7\% Within-1, 6.1\% mean regret@knee, and a cost ratio of 0.051 (94.9\% normalized search-cost savings).

As a representative continuous-depth neural network approach, stochastic differential equation (SDE)-based Bayesian neural networks (BNNs) have attracted considerable attention due to their solid theoretical foundations and strong potential for real-world applications. However, their reliance on numerical SDE solvers inevitably incurs a large number of function evaluations (NFEs), resulting in high computational cost and occasional convergence instability. To address these challenges, we propose a Nesterov-accelerated gradient (NAG) enhanced SDE-BNN model. By integrating NAG into the SDE-BNN framework along with an NFE-dependent residual skip connection, our method accelerates convergence and substantially reduces NFEs during both training and testing. Extensive empirical results show that our model consistently outperforms conventional SDE-BNNs across various tasks, including image classification and sequence modeling, achieving lower NFEs and improved predictive accuracy.

Knowledge distillation, model extraction, and behavior transfer have become central concerns in frontier AI. The main risk is not merely copying, but the possibility that useful capability can be transferred more cheaply than the governance structure that originally accompanied it. This paper presents a public, trade-secret-safe theoretical framework for reducing that asymmetry at the architectural level. The core claim is that distillation becomes less valuable as a shortcut when high-level capability is coupled to internal stability constraints that shape state transitions over time. To formalize this idea, the paper introduces a constraint-coupled reasoning framework with four elements: bounded transition burden, path-load accumulation, dynamically evolving feasible regions, and a capability-stability coupling condition. The paper is intentionally public-safe: it omits proprietary implementation details, training recipes, thresholds, hidden-state instrumentation, deployment procedures, and confidential system design choices. The contribution is therefore theoretical rather than operational. It offers a falsifiable architectural thesis, a clear threat model, and a set of experimentally testable hypotheses for future work on distillation resistance, alignment, and model governance.

System prompt instructions that cooperate in English compete in Spanish, with the same semantic content, but opposite interaction topology. We present instruction-level ablation experiments across four languages and four models showing that this topology inversion is mediated by social register: the imperative mood carries different obligatory force across speech communities, and models trained on multilingual data have learned these conventions. Declarative rewriting of a single instruction block reduces cross-linguistic variance by 81% (p = 0.029, permutation test). Rewriting three of eleven imperative blocks shifts Spanish instruction topology from competitive to cooperative, with spillover effects on unrewritten blocks. These findings suggest that models process instructions as social acts, not technical specifications: "NEVER do X" is an exercise of authority whose force is language-dependent, while "X: disabled" is a factual description that transfers across languages. If register mediates instruction-following at inference time, it plausibly does so during training. We state this as a testable prediction: constitutional AI principles authored in imperative mood may create language-dependent alignment. Corpus: 22 hand-authored probes against a production system prompt decomposed into 56 blocks.

Grokking the delayed transition from memorization to generalization in neural networks remains poorly understood, in part because prior empirical studies confound the roles of architecture, optimization, and regularization. We present a controlled study that systematically disentangles these factors on modular addition (mod 97), with matched and carefully tuned training regimes across models. Our central finding is that grokking dynamics are not primarily determined by architecture, but by interactions between optimization stability and regularization. Specifically, we show: (1) \textbf{depth has a non-monotonic effect}, with depth-4 MLPs consistently failing to grok while depth-8 residual networks recover generalization, demonstrating that depth requires architectural stabilization; (2) \textbf{the apparent gap between Transformers and MLPs largely disappears} (1.11$\times$ delay) under matched hyperparameters, indicating that previously reported differences are largely due to optimizer and regularization confounds; (3) \textbf{activation function effects are regime-dependent}, with GELU up to 4.3$\times$ faster than ReLU only when regularization permits memorization; and (4) \textbf{weight decay is the dominant control parameter}, exhibiting a narrow ``Goldilocks'' regime in which grokking occurs, while too little or too much prevents generalization. Across 3--5 seeds per configuration, these results provide a unified empirical account of grokking as an interaction-driven phenomenon. Our findings challenge architecture-centric interpretations and clarify how optimization and regularization jointly govern delayed generalization.

This paper presents Few TensoRF, a 3D reconstruction framework that combines TensorRF's efficient tensor based representation with FreeNeRF's frequency driven few shot regularization. Using TensorRF to significantly accelerate rendering speed and introducing frequency and occlusion masks, the method improves stability and reconstruction quality under sparse input views. Experiments on the Synthesis NeRF benchmark show that Few TensoRF method improves the average PSNR from 21.45 dB (TensorRF) to 23.70 dB, with the fine tuned version reaching 24.52 dB, while maintaining TensorRF's fast \(\approx10-15\) minute training time. Experiments on the THuman 2.0 dataset further demonstrate competitive performance in human body reconstruction, achieving 27.37 - 34.00 dB with only eight input images. These results highlight Few TensoRF as an efficient and data effective solution for real-time 3D reconstruction across diverse scenes.

Early detection of rice leaf diseases is critical, as rice is a staple crop supporting a substantial share of the world's population. Timely identification of these diseases enables more effective intervention and significantly reduces the risk of large-scale crop losses. However, traditional deep learning models primarily rely on cross entropy loss, which often struggles with high intra-class variance and inter-class similarity, common challenges in plant pathology datasets. To tackle this, we propose a dual-loss framework that combines Center Loss and ArcFace Loss to enhance fine-grained classification of rice leaf diseases. The method is applied into three state-of-the-art backbone architectures: InceptionNetV3, DenseNet201, and EfficientNetB0 trained on the public Rice Leaf Dataset. Our approach achieves significant performance gains, with accuracies of 99.6%, 99.2% and 99.2% respectively. The results demonstrate that angular margin-based and center-based constraints substantially boost the discriminative strength of feature embeddings. In particular, the framework does not require major architectural modifications, making it efficient and practical for real-world deployment in farming environments.

Programming is a core skill in computer science and software engineering (SE), yet identifying and resolving code errors remains challenging for both novice and experienced developers. While Large Language Models (LLMs) have shown remarkable capabilities in natural language understanding and generation tasks, their potential in domain-specific, complex scenarios, such as multi-label classification (MLC) of programming errors, remains underexplored. Recognizing this less-explored area, this study proposes a multi-label error classification (MLEC) framework for source code that leverages fine-tuned LLMs, including CodeT5-base, GraphCodeBERT, CodeT5+, UniXcoder, RoBERTa, PLBART, and CoTexT. These LLMs are integrated with deep learning (DL) architectures such as GRU, LSTM, BiLSTM, and BiLSTM with an additive attention mechanism (BiLSTM-A) to capture both syntactic and semantic features from a real-world student-written Python code error dataset. Extensive experiments across 32 model variants, optimized using Optuna-based hyperparameter tuning, have been evaluated using comprehensive multi-label metrics, including average accuracy, macro and weighted precision, recall, F1-score, exact match accuracy, One-error, Hamming loss, Jaccard similarity, and ROC-AUC (micro, macro, and weighted). Results show that the CodeT5+\_GRU model achieved the strongest performance, with a weighted F1-score of 0.8243, average accuracy of 91.84\%, exact match accuracy of 53.78\%, Hamming loss of 0.0816, and One error of 0.0708. These findings confirm the effectiveness of combining pretrained semantic encoders with efficient recurrent decoders. This work lays the foundation for developing intelligent, scalable tools for automated code feedback, with potential applications in programming education (PE) and broader SE domains.

Failure attribution is essential for diagnosing and improving multi-agent systems (MAS), yet existing benchmarks and methods largely assume a single deterministic root cause for each failure. In practice, MAS failures often admit multiple plausible attributions due to complex inter-agent dependencies and ambiguous execution trajectories. We revisit MAS failure attribution from a multi-perspective standpoint and propose multi-perspective failure attribution, a practical paradigm that explicitly accounts for attribution ambiguity. To support this setting, we introduce MP-Bench, the first benchmark designed for multi-perspective failure attribution in MAS, along with a new evaluation protocol tailored to this paradigm. Through extensive experiments, we find that prior conclusions suggesting LLMs struggle with failure attribution are largely driven by limitations in existing benchmark designs. Our results highlight the necessity of multi-perspective benchmarks and evaluation protocols for realistic and reliable MAS debugging.

The validity of assessments, from large-scale AI benchmarks to human classrooms, depends on the quality of individual items, yet modern evaluation instruments often contain thousands of items with minimal psychometric vetting. We introduce a new family of nonparametric scalability coefficients based on interitem isotonic regression for efficiently detecting globally bad items (e.g., miskeyed, ambiguously worded, or construct-misaligned). The central contribution is the signed isotonic $R^2$, which measures the maximal proportion of variance in one item explainable by a monotone function of another while preserving the direction of association via Kendall's $τ$. Aggregating these pairwise coefficients yields item-level scores that sharply separate problematic items from acceptable ones without assuming linearity or committing to a parametric item response model. We show that the signed isotonic $R^2$ is extremal among monotone predictors (it extracts the strongest possible monotone signal between any two items) and show that this optimality property translates directly into practical screening power. Across three AI benchmark datasets (HS Math, GSM8K, MMLU) and two human assessment datasets, the signed isotonic $R^2$ consistently achieves top-tier AUC for ranking bad items above good ones, outperforming or matching a comprehensive battery of classical test theory, item response theory, and dimensionality-based diagnostics. Crucially, the method remains robust under the small-n/large-p conditions typical of AI evaluation, requires only bivariate monotone fits computable in seconds, and handles mixed item types (binary, ordinal, continuous) without modification. It is a lightweight, model-agnostic filter that can materially reduce the reviewer effort needed to find flawed items in modern large-scale evaluation regimes.

Learning diverse and high-fidelity traffic simulations from human driving demonstrations is crucial for autonomous driving evaluation. The recent next-token prediction (NTP) paradigm, widely adopted in large language models (LLMs), has been applied to traffic simulation and achieves iterative improvements via supervised fine-tuning (SFT). However, such methods limit active exploration of potentially valuable motion tokens, particularly in suboptimal regions. Entropy patterns provide a promising perspective for enabling exploration driven by motion token uncertainty. Motivated by this insight, we propose a novel tokenized traffic simulation policy, R1Sim, which represents an initial attempt to explore reinforcement learning based on motion token entropy patterns, and systematically analyzes the impact of different motion tokens on simulation outcomes. Specifically, we introduce an entropy-guided adaptive sampling mechanism that focuses on previously overlooked motion tokens with high uncertainty yet high potential. We further optimize motion behaviors using Group Relative Policy Optimization (GRPO), guided by a safety-aware reward design. Overall, these components enable a balanced exploration-exploitation trade-off through diverse high-uncertainty sampling and group-wise comparative estimation, resulting in realistic, safe, and diverse multi-agent behaviors. Extensive experiments on the Waymo Sim Agent benchmark demonstrate that R1Sim achieves competitive performance compared to state-of-the-art methods.

Large language model based health agents are increasingly used by health consumers and clinicians to interpret health information and guide health decisions. However, most AI systems in healthcare operate in siloed configurations, supporting individual users rather than the multi-stakeholder relationships central to healthcare. Such use can fragment understanding and exacerbate misalignment among patients, caregivers, and clinicians. We reframe AI not as a standalone assistant, but as a collaborator embedded within multi-party care interactions. Through a clinically validated fictional pediatric chronic kidney disease case study, we show that breakdowns in adherence stem from fragmented situational awareness and misaligned goals, and that siloed use of general-purpose AI tools does little to address these collaboration gaps. We propose a conceptual framework for designing AI collaborators that surface contextual information, reconcile mental models, and scaffold shared understanding while preserving human decision authority.

The rapid advancement of large language models has increasingly blurred the boundary between human-written and AI-generated text, raising societal risks such as misinformation dissemination, authorship ambiguity, and threats to intellectual property rights. These concerns highlight the urgent need for effective and reliable detection methods. While existing training-free approaches often achieve strong performance by aggregating token-level signals into a global score, they typically assume uniform token contributions, making them less robust under short sequences or localized token modifications. To address these limitations, we propose Exons-Detect, a training-free method for AI-generated text detection based on an exon-aware token reweighting perspective. Exons-Detect identifies and amplifies informative exonic tokens by measuring hidden-state discrepancy under a dual-model setting, and computes an interpretable translation score from the resulting importance-weighted token sequence. Empirical evaluations demonstrate that Exons-Detect achieves state-of-the-art detection performance and exhibits strong robustness to adversarial attacks and varying input lengths. In particular, it attains a 2.2\% relative improvement in average AUROC over the strongest prior baseline on DetectRL.

Feature selection is a crucial step in large-scale industrial machine learning systems, directly affecting model accuracy, efficiency, and maintainability. Traditional feature selection methods rely on labeled data and statistical heuristics, making them difficult to apply in production environments where labeled data are limited and multiple operational constraints must be satisfied. To address this, we propose Model Feature Agent (MoFA), a model-driven framework that performs sequential, reasoning-based feature selection using both semantic and quantitative feature information. MoFA incorporates feature definitions, importance scores, correlations, and metadata (e.g., feature groups or types) into structured prompts and selects features through interpretable, constraint-aware reasoning. We evaluate MoFA in three real-world industrial applications: (1) True Interest and Time-Worthiness Prediction, where it improves accuracy while reducing feature group complexity, (2) Value Model Enhancement, where it discovers high-order interaction terms that yield substantial engagement gains in online experiments, and (3) Notification Behavior Prediction, where it selects compact, high-value feature subsets that improve both model accuracy and inference efficiency. Together, these results demonstrate the practicality and effectiveness of LLM-based reasoning for feature selection in real production systems.

Firms that price perishable resources -- airline seats, hotel rooms, seasonal inventory -- now routinely use demand predictions, but these predictions vary widely in quality. Under hard capacity constraints, acting on an inaccurate prediction can irreversibly deplete inventory needed for future periods. We study how prediction uncertainty propagates into dynamic pricing decisions with linear demand, stochastic noise, and finite capacity. A certified demand forecast with known error bound~$ε^0$ specifies where the system should operate: it shifts regret from $O(\sqrt{T})$ to $O(\log T)$ when $ε^0 \lesssim T^{-1/4}$, and we prove this threshold is tight. A misspecified surrogate model -- biased but correlated with true demand -- cannot set prices directly but reduces learning variance by a factor of $(1-ρ^2)$ through control variates. The two mechanisms compose: the forecast determines the regret regime; the surrogate tightens estimation within it. All algorithms rest on a boundary attraction mechanism that stabilizes pricing near degenerate capacity boundaries without requiring non-degeneracy assumptions. Experiments confirm the phase transition threshold, the variance reduction from surrogates, and robustness across problem instances.

Low-field (LF) magnetic resonance imaging (MRI) improves accessibility and reduces costs but generally has lower signal-to-noise ratios and degraded contrast compared to high field (HF) MRI, limiting its clinical utility. Simulating LF MRI from HF MRI enables virtual evaluation of novel imaging devices and development of LF algorithms. Existing low field simulators rely on noise injection and smoothing, which fail to capture the contrast degradation seen in LF acquisitions. To this end, we introduce an end-to-end LF-MRI synthesis framework that learns HF to LF image degradation directly from a small number of paired HF-LF MRIs. Specifically, we introduce a novel HF to LF coordinate-image decoupled neural operator (H2LO) to model the underlying degradation process, and tailor it to capture high-frequency noise textures and image structure. Experimental results in T1w and T2w MRI demonstrate that H2LO produces more faithful simulated low-field images than existing parameterized noise synthesis models and popular image-to-image translation models. Furthermore, it improves performance in downstream image enhancement tasks, showcasing its potential to enhance LF MRI diagnostic capabilities.

Understanding why a large language model (LLM) is uncertain about the response is important for their reliable deployment. Current approaches, which either provide a single uncertainty score or rely on the classical aleatoric-epistemic dichotomy, fail to offer actionable insights for improving the generative model. Recent studies have also shown that such methods are not enough for understanding uncertainty in LLMs. In this work, we advocate for an uncertainty decomposition framework that dissects LLM uncertainty into three distinct semantic components: (i) input ambiguity, arising from ambiguous prompts; (ii) knowledge gaps, caused by insufficient parametric evidence; and (iii) decoding randomness, stemming from stochastic sampling. Through a series of experiments we demonstrate that the dominance of these components can shift across model size and task. Our framework provides a better understanding to audit LLM reliability and detect hallucinations, paving the way for targeted interventions and more trustworthy systems.

Despite significant progress in text-to-image generation, aligning outputs with complex prompts remains challenging, particularly for fine-grained semantics and spatial relations. This difficulty stems from the feed-forward nature of generation, which requires anticipating alignment without fully understanding the output. In contrast, evaluating generated images is more tractable. Motivated by this asymmetry, we propose xLARD, a self-correcting framework that uses multimodal large language models to guide generation through Explainable LAtent RewarDs. xLARD introduces a lightweight corrector that refines latent representations based on structured feedback from model-generated references. A key component is a differentiable mapping from latent edits to interpretable reward signals, enabling continuous latent-level guidance from non-differentiable image-level evaluations. This mechanism allows the model to understand, assess, and correct itself during generation. Experiments across diverse generation and editing tasks show that xLARD improves semantic alignment and visual fidelity while maintaining generative priors. Code is available at https://yinyiluo.github.io/xLARD/.

Modern computational advertising platforms typically rely on recommendation systems to predict user responses, such as click-through rates, conversion rates, and other optimization events. To support a wide variety of product surfaces and advertiser goals, these platforms frequently maintain an extensive ecosystem of machine learning (ML) models. However, operating at this scale creates significant development and efficiency challenges. Substantial engineering effort is required to regularly refresh ML models and propagate new techniques, which results in long latencies when deploying ML innovations across the ecosystem. We present a large-scale empirical study comparing model performance, efficiency, and ML technique propagation between a standardized model-building approach and independent per-model optimization in recommendation systems. To facilitate this standardization, we propose the Standard Model Template (SMT) -- a framework that generates high-performance models adaptable to diverse data distributions and optimization events. By utilizing standardized, composable ML model components, SMT reduces technique propagation complexity from $O(n \cdot 2^k)$ to $O(n + k)$ where $n$ is the number of models and $k$ the number of techniques. Evaluating an extensive suite of models over four global development cycles within Meta's production ads ranking ecosystem, our results demonstrate: (1) a 0.63% average improvement in cross-entropy at neutral serving capacity, (2) a 92% reduction in per-model iteration engineering time, and (3) a $6.3\times$ increase in technique-model pair adoption throughput. These findings challenge the conventional wisdom that diverse optimization goals inherently require diversified ML model design.

Assessing student handwritten scratchwork is crucial for personalized educational feedback but presents unique challenges due to diverse handwriting, complex layouts, and varied problem-solving approaches. Existing educational NLP primarily focuses on textual responses and neglects the complexity and multimodality inherent in authentic handwritten scratchwork. Current multimodal large language models (MLLMs) excel at visual reasoning but typically adopt an "examinee perspective", prioritizing generating correct answers rather than diagnosing student errors. To bridge these gaps, we introduce ScratchMath, a novel benchmark specifically designed for explaining and classifying errors in authentic handwritten mathematics scratchwork. Our dataset comprises 1,720 mathematics samples from Chinese primary and middle school students, supporting two key tasks: Error Cause Explanation (ECE) and Error Cause Classification (ECC), with seven defined error types. The dataset is meticulously annotated through rigorous human-machine collaborative approaches involving multiple stages of expert labeling, review, and verification. We systematically evaluate 16 leading MLLMs on ScratchMath, revealing significant performance gaps relative to human experts, especially in visual recognition and logical reasoning. Proprietary models notably outperform open-source models, with large reasoning models showing strong potential for error explanation. All evaluation data and frameworks are publicly available to facilitate further research.

Machine Translation (MT) and Quality Estimation (QE) perform well in general domains but degrade under domain mismatch. This dissertation studies how to adapt MT and QE systems to specialized domains through a set of data-focused contributions. Chapter 2 presents a similarity-based data selection method for MT. Small, targeted in-domain subsets outperform much larger generic datasets and reach strong translation quality at lower computational cost. Chapter 3 introduces a staged QE training pipeline that combines domain adaptation with lightweight data augmentation. The method improves performance across domains, languages, and resource settings, including zero-shot and cross-lingual cases. Chapter 4 studies the role of subword tokenization and vocabulary in fine-tuning. Aligned tokenization-vocabulary setups lead to stable training and better translation quality, while mismatched configurations reduce performance. Chapter 5 proposes a QE-guided in-context learning method for large language models. QE models select examples that improve translation quality without parameter updates and outperform standard retrieval methods. The approach also supports a reference-free setup, reducing reliance on a single reference set. These results show that domain adaptation depends on data selection, representation, and efficient adaptation strategies. The dissertation provides methods for building MT and QE systems that perform reliably in domain-specific settings.

This paper provides some of the first large-scale descriptive evidence on how consumers adopt and use platform-embedded shopping AI in e-commerce. Using data on 31 million users of Ctrip, China's largest online travel platform, we study "Wendao," an LLM-based AI assistant integrated into the platform. We document three empirical regularities. First, adoption is highest among older consumers, female users, and highly engaged existing users, reversing the younger, male-dominated profile commonly documented for general-purpose AI tools. Second, AI chat appears in the same broad phase of the purchase journey as traditional search and well before order placement; among journeys containing both chat and search, the most common pattern is interleaving, with users moving back and forth between the two modalities. Third, consumers disproportionately use the assistant for exploratory, hard-to-keyword tasks: attraction queries account for 42% of observed chat requests, and chat intent varies systematically with both the timing of chat relative to search and the category of products later purchased within the same journey. These findings suggest that embedded shopping AI functions less as a substitute for conventional search than as a complementary interface for exploratory product discovery in e-commerce.

Large language models (LLMs) have shown strong performance on automated software engineering tasks, yet existing benchmarks focus primarily on general-purpose libraries or web applications, leaving mobile application development largely unexplored despite its strict platform constraints, framework-driven lifecycles, and complex platform API interactions. We introduce MobileDev-Bench, a benchmark comprising 384 real-world issue-resolution tasks collected from 18 production mobile applications spanning Android Native (Java/Kotlin), React Native (TypeScript), and Flutter (Dart). Each task pairs an authentic developer-reported issue with executable test patches, enabling fully automated validation of model-generated fixes within mobile build environments. The benchmark exhibits substantial patch complexity: fixes modify 12.5 files and 324.9 lines on average, and 35.7% of instances require coordinated changes across multiple artifact types, such as source and manifest files. Evaluation of four state-of-the-art code-capable LLMs, GPT- 5.2, Claude Sonnet 4.5, Gemini Flash 2.5, and Qwen3-Coder, yields low end-to-end resolution rates of 3.39%-5.21%, revealing significant performance gaps compared to prior benchmarks. Further analysis reveals systematic failure modes, with fault localization across multi-file and multi-artifact changes emerging as the primary bottleneck.

This paper introduces \textbf{FinMCP-Bench}, a novel benchmark for evaluating large language models (LLMs) in solving real-world financial problems through tool invocation of financial model context protocols. FinMCP-Bench contains 613 samples spanning 10 main scenarios and 33 sub-scenarios, featuring both real and synthetic user queries to ensure diversity and authenticity. It incorporates 65 real financial MCPs and three types of samples, single tool, multi-tool, and multi-turn, allowing evaluation of models across different levels of task complexity. Using this benchmark, we systematically assess a range of mainstream LLMs and propose metrics that explicitly measure tool invocation accuracy and reasoning capabilities. FinMCP-Bench provides a standardized, practical, and challenging testbed for advancing research on financial LLM agents.

Vision-Language-Action (VLA) models excel in robotic manipulation but suffer from significant inference latency due to processing dense visual tokens. Existing token reduction methods predominantly rely on attention magnitude as a static selection. In this work, we challenge this assumption, revealing that high-attention tokens are task-dependent and can even degrade policy performance. To address this, we introduce \textbf{TIES} (\textbf{T}au-guided \textbf{I}nter-layer \textbf{E}fficient \textbf{S}election), a dynamic framework guided by inter-layer token ranking consistency. By adaptively balancing attention magnitude with ranking consistency, TIES ensures robust token selection without requiring additional training. On the CogACT + SIMPLER benchmark, TIES improves average success rates by 6\% while reducing token usage by 78\%, and demonstrate strong generalization across diverse decoders and benchmarks.

This paper introduces the design and development of a framework that integrates a large language model (LLM) with a retrieval-augmented generation (RAG) approach leveraging both a knowledge graph and user interaction history. The framework is incorporated into a previously developed adaptive learning support system to assess learners' code, generate formative feedback, and recommend exercises. Moerover, this study examines learner preferences across three instructional modes; adaptive, Generative AI (GenAI), and hybrid GenAI-adaptive. An experimental study was conducted to compare the learning performance and perception of the learners, and the effectiveness of these three modes using four key log features derived from 4956 code submissions across all experimental groups. The analysis results show that learners receiving feedback from GenAI modes had significantly more correct code and fewer code submissions missing essential programming logic than those receiving feedback from adaptive mode. In particular, the hybrid GenAI-adaptive mode achieved the highest number of correct submissions and the fewest incorrect or incomplete attempts, outperforming both the adaptive-only and GenAI-only modes. Questionnaire responses further indicated that GenAI-generated feedback was widely perceived as helpful, while all modes were rated positively for ease of use and usefulness. These results suggest that the hybrid GenAI-adaptive mode outperforms the other two modes across all measured log features.

Human trajectory forecasting is important for intelligent multimedia systems operating in visually complex environments, such as autonomous driving and crowd surveillance. Although Conditional Flow Matching (CFM) has shown strong ability in modeling trajectory distributions from spatio-temporal observations, existing approaches still focus primarily on supervised fitting, which may leave social norms and scene constraints insufficiently reflected in generated trajectories. To address this issue, we propose TIGFlow-GRPO, a two-stage generative framework that aligns flow-based trajectory generation with behavioral rules. In the first stage, we build a CFM-based predictor with a Trajectory-Interaction-Graph (TIG) module to model fine-grained visual-spatial interactions and strengthen context encoding. This stage captures both agent-agent and agent-scene relations more effectively, providing more informative conditional features for subsequent alignment. In the second stage, we perform Flow-GRPO post-training,where deterministic flow rollout is reformulated as stochastic ODE-to-SDE sampling to enable trajectory exploration, and a composite reward combines view-aware social compliance with map-aware physical feasibility. By evaluating trajectories explored through SDE rollout, GRPO progressively steers multimodal predictions toward behaviorally plausible futures. Experiments on the ETH/UCY and SDD datasets show that TIGFlow-GRPO improves forecasting accuracy and long-horizon stability while generating trajectories that are more socially compliant and physically feasible. These results suggest that the proposed framework provides an effective way to connect flow-based trajectory modeling with behavior-aware alignment in dynamic multimedia environments.

We propose Context-aware Video-text Alignment (CVA), a novel framework to address a significant challenge in video temporal grounding: achieving temporally sensitive video-text alignment that remains robust to irrelevant background context. Our framework is built on three key components. First, we propose Query-aware Context Diversification (QCD), a new data augmentation strategy that ensures only semantically unrelated content is mixed in. It builds a video-text similarity-based pool of replacement clips to simulate diverse contexts while preventing the ``false negative" caused by query-agnostic mixing. Second, we introduce the Context-invariant Boundary Discrimination (CBD) loss, a contrastive loss that enforces semantic consistency at challenging temporal boundaries, making their representations robust to contextual shifts and hard negatives. Third, we introduce the Context-enhanced Transformer Encoder (CTE), a hierarchical architecture that combines windowed self-attention and bidirectional cross-attention with learnable queries to capture multi-scale temporal context. Through the synergy of these data-centric and architectural enhancements, CVA achieves state-of-the-art performance on major VTG benchmarks, including QVHighlights and Charades-STA. Notably, our method achieves a significant improvement of approximately 5 points in Recall@1 (R1) scores over state-of-the-art methods, highlighting its effectiveness in mitigating false negatives.

The growing prominence of cryptocurrencies has triggered widespread public engagement and increased speculative activity, particularly on social media platforms. This study introduces a novel classification framework for identifying predictive statements in cryptocurrency-related tweets, focusing on five popular cryptocurrencies: Cardano, Matic, Binance, Ripple, and Fantom. The classification process is divided into two stages: Task 1 involves binary classification to distinguish between Predictive and Non-Predictive statements. Tweets identified as Predictive proceed to Task 2, where they are further categorized as Incremental, Decremental, or Neutral. To build a robust dataset, we combined manual and GPT-based annotation methods and utilized SenticNet to extract emotion features corresponding to each prediction category. To address class imbalance, GPT-generated paraphrasing was employed for data augmentation. We evaluated a wide range of machine learning, deep learning, and transformer-based models across both tasks. The results show that GPT-based balancing significantly enhanced model performance, with transformer models achieving the highest F1-score in Task 1, while traditional machine learning models performed best in Task 2. Furthermore, our emotion analysis revealed distinct emotional patterns associated with each prediction category across the different cryptocurrencies.

Understanding and quantifying uncertainty in large language model (LLM) outputs is critical for reliable deployment. However, traditional evaluation approaches provide limited insight into model confidence at individual token positions during generation. To address this issue, we introduce LogitScope, a lightweight framework for analyzing LLM uncertainty through token-level information metrics computed from probability distributions. By measuring metrics such as entropy and varentropy at each generation step, LogitScope reveals patterns in model confidence, identifies potential hallucinations, and exposes decision points where models exhibit high uncertainty, all without requiring labeled data or semantic interpretation. We demonstrate LogitScope's utility across diverse applications including uncertainty quantification, model behavior analysis, and production monitoring. The framework is model-agnostic, computationally efficient through lazy evaluation, and compatible with any HuggingFace model, enabling both researchers and practitioners to inspect LLM behavior during inference.

Semantic search in retrieval-augmented generation (RAG) systems is often insufficient for complex information needs, particularly when relevant evidence is scattered across multiple sources. Prior approaches to this problem include agentic retrieval strategies, which expand the semantic search space by generating additional queries. However, these methods do not fully leverage the organizational structure of the data and instead rely on iterative exploration, which can lead to inefficient retrieval. Another class of approaches employs knowledge graphs to model non-semantic relationships through graph edges. Although effective in capturing richer proximities, such methods incur significant maintenance costs and are often incompatible with the vector stores used in most production systems. To address these limitations, we propose GraphER, a graph-based enrichment and reranking method that captures multiple forms of proximity beyond semantic similarity. GraphER independently enriches data objects during offline indexing and performs graph-based reranking over candidate objects at query time. This design does not require a knowledge graph, allowing GraphER to integrate seamlessly with standard vector stores. In addition, GraphER is retriever-agnostic and introduces negligible latency overhead. Experiments on multiple retrieval benchmarks demonstrate the effectiveness of the proposed approach.

Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantially larger per-sequence extraction mass, and patterns in how near-verbatim extraction risk manifests across model sizes and types of text.

Deploying neural networks on microcontrollers is constrained by kilobytes of flash and SRAM, where 1x1 pointwise (PW) mixers often dominate memory even after INT8 quantization across vision, audio, and wearable sensing. We present HYPER-TINYPW, a compression-as-generation approach that replaces most stored PW weights with generated weights: a shared micro-MLP synthesizes PW kernels once at load time from tiny per-layer codes, caches them, and executes them with standard integer operators. This preserves commodity MCU runtimes and adds only a one-off synthesis cost; steady-state latency and energy match INT8 separable CNN baselines. Enforcing a shared latent basis across layers removes cross-layer redundancy, while keeping PW1 in INT8 stabilizes early, morphology-sensitive mixing. We contribute (i) TinyML-faithful packed-byte accounting covering generator, heads/factorization, codes, kept PW1, and backbone; (ii) a unified evaluation with validation-tuned t* and bootstrap confidence intervals; and (iii) a deployability analysis covering integer-only inference and boot versus lazy synthesis. On three ECG benchmarks (Apnea-ECG, PTB-XL, MIT-BIH), HYPER-TINYPW shifts the macro-F1 versus flash Pareto frontier: at about 225 kB it matches a roughly 1.4 MB CNN while being 6.31x smaller (84.15% fewer bytes), retaining at least 95% of large-model macro-F1. Under 32-64 kB budgets it sustains balanced detection where compact baselines degrade. The mechanism applies broadly to other 1D biosignals, on-device speech, and embedded sensing tasks where per-layer redundancy dominates, indicating a wider role for compression-as-generation in resource-constrained ML systems. Beyond ECG, HYPER-TINYPW transfers to TinyML audio: on Speech Commands it reaches 96.2% test accuracy (98.2% best validation), supporting broader applicability to embedded sensing workloads where repeated linear mixers dominate memory.

Artificial intelligence is transforming mathematics at a speed and scale that demand active engagement from the mathematical community. We examine five areas where this transformation is particularly pressing: values, practice, teaching, technology, and ethics. We offer recommendations on safeguarding our intellectual autonomy, rethinking our practice, broadening curricula, building academically oriented infrastructure, and developing shared ethical principles - with the aim of ensuring that the future of mathematics is shaped by the community itself.

Coordinating teams of aerial robots in cluttered three-dimensional (3D) environments requires a principled integration of discrete mission planning-deciding which robot serves which goals and in what order -- with continuous-time trajectory synthesis that enforces collision avoidance and dynamic feasibility. This paper introduces IMD-TAPP (Integrated Multi-Drone Task Allocation and Path Planning), an end-to-end framework that jointly addresses multi-goal allocation, tour sequencing, and safe trajectory generation for quadrotor teams operating in obstacle-rich spaces. IMD--TAPP first discretizes the workspace into a 3D navigation graph and computes obstacle-aware robot-to-goal and goal-to-goal travel costs via graph-search-based pathfinding. These costs are then embedded within an Injected Particle Swarm Optimization (IPSO) scheme, guided by multiple linear assignment, to efficiently explore coupled assignment/ordering alternatives and to minimize mission makespan. Finally, the resulting waypoint tours are transformed into time-parameterized minimum-snap trajectories through a generation-and-optimization routine equipped with iterative validation of obstacle clearance and inter-robot separation, triggering re-planning when safety margins are violated. Extensive MATLAB simulations across cluttered 3D scenarios demonstrate that IMD--TAPP consistently produces dynamically feasible, collision-free trajectories while achieving competitive completion times. In a representative case study with two drones serving multiple goals, the proposed approach attains a minimum mission time of 136~s while maintaining the required safety constraints throughout execution.

We prove that platform-deterministic inference is necessary and sufficient for trustworthy AI. We formalize this as the Determinism Thesis and introduce trust entropy to quantify the cost of non-determinism, proving that verification failure probability equals 1 - 2^{-H_T} exactly. We prove a Determinism-Verification Collapse: verification under determinism requires O(1) hash comparison; without it, the verifier faces an intractable membership problem. IEEE 754 floating-point arithmetic fundamentally violates the determinism requirement. We resolve this by constructing a pure integer inference engine that achieves bitwise identical output across ARM and x86. In 82 cross-architecture tests on models up to 6.7B parameters, we observe zero hash mismatches. Four geographically distributed nodes produce identical outputs, verified by 356 on-chain attestation transactions. Every major trust property of AI systems (fairness, robustness, privacy, safety, alignment) presupposes platform determinism. Our system, 99,000 lines of Rust deployed across three continents, establishes that AI trust is a question of arithmetic.

We present a Sovereign AI architecture for clinical triage in which all inference is performed on-device and inbound data is delivered via a physically unidirectional channel, implemented using receive-only broadcast infrastructure or certified hardware data diodes, with no return path to any external network. This design removes the network-mediated attack surface by construction, rather than attempting to secure it through software controls. The system performs conversational symptom intake, integrates device-captured vitals, and produces structured, triage-aligned clinical records at the point of care. We formalize the security properties of receiver-side unidirectionality and show that the architecture is transport-agnostic across broadcast and diode-enforced deployments. We further analyze threat models, enforcement mechanisms, and deployment configurations, demonstrating how physical one-way data flow enables high-assurance operation in both resource-constrained and high-risk environments. This work positions physically unidirectional channels as a foundational primitive for sovereign, on-device clinical intelligence at the front door of care.

This paper describes LogSigma, our system for SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis (DimABSA). Unlike traditional Aspect-Based Sentiment Analysis (ABSA), which predicts discrete sentiment labels, DimABSA requires predicting continuous Valence and Arousal (VA) scores on a 1-9 scale. A central challenge is that Valence and Arousal differ in prediction difficulty across languages and domains. We address this using learned homoscedastic uncertainty, where the model learns task-specific log-variance parameters to automatically balance each regression objective during training. Combined with language-specific encoders and multi-seed ensembling, LogSigma achieves 1st place on five datasets across both tracks. The learned variance weights vary substantially across languages due to differing Valence-Arousal difficulty profiles-from 0.66x for German to 2.18x for English-demonstrating that optimal task balancing is language-dependent and cannot be determined a priori.

Spiking Neural Networks (SNNs) offer inherent advantages for low-power inference through sparse, event-driven computation. However, the theoretical energy benefits of SNNs are often decoupled from real hardware performance due to the opaque relationship between training-time choices and inference-time sparsity. While prior work has focused on weight pruning and compression, the role of training hyperparameters -- specifically surrogate gradient functions and neuron model configurations -- in shaping hardware-level activation sparsity remains underexplored. This paper presents a workload characterization study quantifying the sensitivity of hardware latency to SNN hyperparameters. We decouple the impact of surrogate gradient functions (e.g., Fast Sigmoid, Spike Rate Escape) and neuron models (LIF, Lapicque) on classification accuracy and inference efficiency across three event-based vision datasets: DVS128-Gesture, N-MNIST, and DVS-CIFAR10. Our analysis reveals that standard accuracy metrics are poor predictors of hardware efficiency. While Fast Sigmoid achieves the highest accuracy on DVS-CIFAR10, Spike Rate Escape reduces inference latency by up to 12.2% on DVS128-Gesture with minimal accuracy trade-offs. We also demonstrate that neuron model selection is as critical as parameter tuning; transitioning from LIF to Lapicque neurons yields up to 28% latency reduction. We validate on a custom cycle-accurate FPGA-based SNN instrumentation platform, showing that sparsity-aware hyperparameter selection can improve accuracy by 9.1% and latency by over 2x compared to baselines. These findings establish a methodology for predicting hardware behavior from training parameters. The RTL and reproducibility artifacts are at https://zenodo.org/records/18893738.

We investigate machine learning approaches for optimizing real-time staffing decisions in semi-automated warehouse sortation systems. Operational decision-making can be supported at different levels of abstraction, with different trade-offs. We evaluate two approaches, each in a matching simulation environment. First, we train custom Transformer-based policies using offline reinforcement learning on detailed historical state representations, achieving a 2.4% throughput improvement over historical baselines in learned simulators. In high-volume warehouse operations, improvements of this size translate to significant savings. Second, we explore LLMs operating on abstracted, human-readable state descriptions. These are a natural fit for decisions that warehouse managers make using high-level operational summaries. We systematically compare prompting techniques, automatic prompt optimization, and fine-tuning strategies. While prompting alone proves insufficient, supervised fine-tuning combined with Direct Preference Optimization on simulator-generated preferences achieves performance that matches or slightly exceeds historical baselines in a hand-crafted simulator. Our findings demonstrate that both approaches offer viable paths toward AI-assisted operational decision-making. Offline RL excels with task-specific architectures. LLMs support human-readable inputs and can be combined with an iterative feedback loop that can incorporate manager preferences.

Open source software (OSS) sustainability depends not only on code contributions but also on governance structures that define who decides, who acts, and how responsibility is distributed. We lack systematic empirical evidence of how projects formally codify roles and authority in written artifacts. This paper investigates how OSS projects define and structure governance through their GOVERNANCE.md files and related documents. We analyze governance as an institutional infrastructure, a set of explicit rules that shape participation, decision rights, and community memory. We used Institutional Grammar to extract and formalize role definitions from repositories hosted on GitHub. We decompose each role into scope, privileges, obligations, and life-cycle rules to compare role structures across communities. Our results show that although OSS projects use a stable set of titles, identical titles carry different responsibilities, and different labels describe similar functions, which we call role drift. Still, we observed that a few actors sometimes accumulate technical, managerial, and community duties. %This creates the Maintainer Paradox: those who enable broad participation simultaneously become governance bottlenecks. By understanding authority and responsibilities in OSS, our findings inform researchers and practitioners on the importance of designing clearer roles, distributing work, and reducing leadership overload to support healthier and more sustainable communities.

Generative artificial intelligence (AI) tools can now help people perform complex data science tasks regardless of their expertise. While these tools have great potential to help more people work with data, their end-to-end approach does not support users in evaluating alternative approaches and reformulating problems, both critical to solving open-ended tasks in high-stakes domains. In this paper, we reflect on two AI data science systems designed for the medical setting and how they function as tools for thought. We find that success in these systems was driven by constructing AI workflows around intentionally-designed intermediate artifacts, such as readable query languages, concept definitions, or input-output examples. Despite opaqueness in other parts of the AI process, these intermediates helped users reason about important analytical choices, refine their initial questions, and contribute their unique knowledge. We invite the HCI community to consider when and how intermediate artifacts should be designed to promote effective data science thinking.

The physical world is not merely visual; it is governed by rigorous structural and procedural constraints. Yet, the evaluation of vision-language models (VLMs) remains heavily skewed toward perceptual realism, prioritizing the generation of visually plausible 3D layouts, shapes, and appearances. Current benchmarks rarely test whether models grasp the step-by-step processes and physical dependencies required to actually build these artifacts, a capability essential for automating design-to-construction pipelines. To address this, we introduce DreamHouse, a novel benchmark for physical generative reasoning: the capacity to synthesize artifacts that concurrently satisfy geometric, structural, constructability, and code-compliance constraints. We ground this benchmark in residential timber-frame construction, a domain with fully codified engineering standards and objectively verifiable correctness. We curate over 26,000 structures spanning 13 architectural styles, ach verified to construction-document standards (LOD 350) and develop a deterministic 10-test structural validation framework. Unlike static benchmarks that assess only final outputs, DreamHouse supports iterative agentic interaction. Models observe intermediate build states, generate construction actions, and receive structured environmental feedback, enabling a fine-grained evaluation of planning, structural reasoning, and self-correction. Extensive experiments with state-of-the-art VLMs reveal substantial capability gaps that are largely invisible on existing leaderboards. These findings establish physical validity as a critical evaluation axis orthogonal to visual realism, highlighting physical generative reasoning as a distinct and underdeveloped frontier in multimodal intelligence. Available at https://luluyuyuyang.github.io/dreamhouse

As machine learning (ML) systems expand in both scale and functionality, the security landscape has become increasingly complex, with a proliferation of attacks and defenses. However, existing studies largely treat these threats in isolation, lacking a coherent framework to expose their shared principles and interdependencies. This fragmented view hinders systematic understanding and limits the design of comprehensive defenses. Crucially, the two foundational assets of ML -- \textbf{data} and \textbf{models} -- are no longer independent; vulnerabilities in one directly compromise the other. The absence of a holistic framework leaves open questions about how these bidirectional risks propagate across the ML pipeline. To address this critical gap, we propose a \emph{unified closed-loop threat taxonomy} that explicitly frames model-data interactions along four directional axes. Our framework offers a principled lens for analyzing and defending foundation models. The resulting four classes of security threats represent distinct but interrelated categories of attacks: (1) Data$\rightarrow$Data (D$\rightarrow$D): including \emph{data decryption attacks and watermark removal attacks}; (2) Data$\rightarrow$Model (D$\rightarrow$M): including \emph{poisoning, harmful fine-tuning attacks, and jailbreak attacks}; (3) Model$\rightarrow$Data (M$\rightarrow$D): including \emph{model inversion, membership inference attacks, and training data extraction attacks}; (4) Model$\rightarrow$Model (M$\rightarrow$M): including \emph{model extraction attacks}. Our unified framework elucidates the underlying connections among these security threats and establishes a foundation for developing scalable, transferable, and cross-modal security strategies, particularly within the landscape of foundation models.

Emergency response systems generate data from many agencies and systems. In practice, correlating and updating this information across sources in a way that aligns with Next Generation 9-1-1 data standards remains challenging. Ideally, this data should be treated as a continuous stream of operational updates, where new facts are integrated immediately to provide a timely and unified view of an evolving incident. This paper presents SentinelAI, a data integration and standardization framework for transforming emergency communications into standardized, machine-readable datasets that support integration, composite incident construction, and cross-source reasoning. SentinelAI implements a scalable processing pipeline composed of specialized agents. The EIDO Agent ingests raw communications and produces NENA-compliant Emergency Incident Data Object JSON.

Ethical debates in AI have primarily focused on back-end issues such as data governance, model training, and algorithmic decision-making. Less attention has been paid to the ethical significance of front-end design choices, such as the interaction and representation-based elements through which users interact with AI systems. This gap is particularly significant for Conversational User Interfaces (CUI) based on Natural Language Processing (NLP) systems, where humanizing design elements such as dialogue-based interaction, emotive language, personality modes, and anthropomorphic metaphors are increasingly prevalent. This work argues that humanization in AI front-end design is a value-driven choice that profoundly shapes users' mental models, trust calibration, and behavioral responses. Drawing on research in human-computer interaction (HCI), conversational AI, and value-sensitive design, we examine how interfaces can play a central role in misaligning user expectations, fostering misplaced trust, and subtly undermining user autonomy, especially in vulnerable contexts. To ground this analysis, we discuss two AI systems developed by Chayn, a nonprofit organization supporting survivors of gender-based violence. Chayn is extremely cautious when building AI that interacts with or impacts survivors by operationalizing their trauma-informed design principles. This Chayn case study illustrates how ethical considerations can motivate principled restraint in interface design, challenging engagement-based norms in contemporary AI products. We argue that ethical front-end AI design is a form of procedural ethics, enacted through interaction choices rather than embedded solely in system logic.

Fire safety consists of a complex pipeline, and it is a very important topic of concern. One of its frontal parts are the smoke detectors, which are supposed to provide an alarm prior to a massive fire appears. As they are often difficult to reach due to high ceilings or problematic locations, an automatic inspection system would be very beneficial as it could allow faster revisions, prevent workers from dangerous work in heights, and make the whole process cheaper. In this study, we present the smoke detector recognition part of the automatic inspection system, which could easily be integrated to the drone system. As part of our research, we compare two popular convolutional-based object detectors YOLOv11 and SSD widely used on embedded devices together with the state-of-the-art transformer-based RT-DETRv2 with the backbones of different sizes. Due to a complicated way of collecting a sufficient amount of data for training in the real-world environment, we also compare several training strategies using the real and semi-synthetic data together with various augmentation methods. To achieve a robust testing, all models were evaluated on two test datasets with an expected and difficult appearance of the smoke detectors including motion blur, small resolution, or not complete objects. The best performing detector is the YOLOv11n, which reaches the average mAP@0.5 score of 0.884. Our code, pretrained models and dataset are publicly available.

Preference learning methods, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO), rely on pairwise human judgments, yet little is known about the cognitive processes underlying these judgments. We investigate whether eye-tracking can reveal preference formation during pairwise AI-generated image evaluation. Thirty participants completed 1,800 trials while their gaze was recorded. We replicated the gaze cascade effect, with gaze shifting toward chosen images approximately one second before the decision. Cascade dynamics were consistent across confidence levels. Gaze features predicted binary choice (68% accuracy), with chosen images receiving more dwell time, fixations, and revisits. Gaze transitions distinguished high-confidence from uncertain decisions (66% accuracy), with low-confidence trials showing more image switches per second. These results show that gaze patterns predict both choice and confidence in pairwise image evaluations, suggesting that eye-tracking provides implicit signals relevant to the quality of preference annotations.

Recent advances in multimodal large language models enable new possibilities for image-based decision support. However, their reliability and operational trade-offs in neuroimaging remain insufficiently understood. We present a comprehensive benchmarking study of vision-enabled large language models for 2D neuroimaging using curated MRI and CT datasets covering multiple sclerosis, stroke, brain tumors, other abnormalities, and normal controls. Models are required to generate multiple outputs simultaneously, including diagnosis, diagnosis subtype, imaging modality, specialized sequence, and anatomical plane. Performance is evaluated across four directions: discriminative classification with abstention, calibration, structured-output validity, and computational efficiency. A multi-phase framework ensures fair comparison while controlling for selection bias. Across twenty frontier multimodal models, the results show that technical imaging attributes such as modality and plane are nearly solved, whereas diagnostic reasoning, especially subtype prediction, remains challenging. Tumor classification emerges as the most reliable task, stroke is moderately solvable, while multiple sclerosis and rare abnormalities remain difficult. Few-shot prompting improves performance for several models but increases token usage, latency, and cost. Gemini-2.5-Pro and GPT-5-Chat achieve the strongest overall diagnostic performance, while Gemini-2.5-Flash offers the best efficiency-performance trade-off. Among open-weight architectures, MedGemma-1.5-4B demonstrates the most promising results, as under few-shot prompting, it approaches the zero-shot performance of several proprietary models, while maintaining perfect structured output. These findings provide practical insights into performance, reliability, and efficiency trade-offs, supporting standardized evaluation of multimodal LLMs in neuroimaging.

Given a question, a language model (LM) implicitly encodes a distribution over possible answers. In practice, post-training procedures for LMs often collapse this distribution onto a single dominant mode. While this is generally not a problem for benchmark-style evaluations that assume one correct answer, many real-world tasks inherently involve multiple valid answers or irreducible uncertainty. Examples include medical diagnosis, ambiguous question answering, and settings with incomplete information. In these cases, we would like LMs to generate multiple plausible hypotheses, ideally with confidence estimates for each one, and without computationally intensive repeated sampling to generate non-modal answers. This paper describes a multi-answer reinforcement learning approach for training LMs to perform distributional reasoning over multiple answers during inference. We modify the RL objective to enable models to explicitly generate multiple candidate answers in a single forward pass, internalizing aspects of inference-time search into the model's generative process. Across question-answering, medical diagnostic, and coding benchmarks, we observe improved diversity, coverage, and set-level calibration scores compared to single answer trained baselines. Models trained with our approach require fewer tokens to generate multiple answers than competing approaches. On coding tasks, they are also substantially more accurate. These results position multi-answer RL as a principled and compute-efficient alternative to inference-time scaling procedures such as best-of-k. Code and more information can be found at https://multi-answer-rl.github.io/.

Spacecraft development costs remain high despite falling launch costs, in part because Model-Based Systems Engineering (MBSE) tools carry the complexity of the object-oriented programming paradigm: tightly coupled data and logic, mutable state, and rigid class hierarchies that resist integration with discipline-specific analysis tools. This paper presents a data-oriented approach to MBSE that adapts the Entity-Component-System (ECS) architecture from the video game industry. Design data is stored as immutable, format-agnostic components in a generic data system; stateless analysis functions operate on this data through templates and containerized tools within a continuous integration pipeline. A prototype implementation, VVERDAD (https://github.com/VisVivaSpace/vverdad-prototype), demonstrates the approach on an example interplanetary mission concept, showing how data-oriented principles can reduce deployment complexity, simplify testing, and preserve the traceability benefits of document-based systems engineering.

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, methods such as GRPO and DAPO suffer from substantial computational cost, since they rely on sampling many rollouts for each prompt. Moreover, in RLVR the relative advantage is often sparse: many samples become nearly all-correct or all-incorrect, yielding low within-group reward variance and thus weak learning signals. In this paper, we introduce arrol (Accelerating RLVR via online Rollout Pruning), an online rollout pruning method that prunes rollouts during generation while explicitly steering the surviving ones more correctness-balanced to enhance learning signals. Specifically, arrol trains a lightweight quality head on-the-fly to predict the success probability of partial rollouts and uses it to make early pruning decisions. The learned quality head can further weigh candidates to improve inference accuracy during test-time scaling. To improve efficiency, we present a system design that prunes rollouts inside the inference engine and re-batches the remaining ones for log-probability computation and policy updates. Across GRPO and DAPO on Qwen-3 and LLaMA-3.2 models (1B-8B), arrol improves average accuracy by +2.30 to +2.99 while achieving up to 1.7x training speedup, and yielding up to +8.33 additional gains in average accuracy in test-time scaling. The code is available at https://github.com/Hsu1023/ARRoL.

We propose a general framework to extend Flow Matching to homogeneous spaces, i.e. quotients of Lie groups. Our approach reformulates the problem as a flow matching task on the underlying Lie group by lifting the data distributions. This strategy avoids the potentially complicated geometry of homogeneous spaces by working directly on Lie groups, which in turn enables us reduce the problem to a Euclidean flow matching task on Lie algebras. In contrast to Riemannian Flow Matching, our method eliminates the need to define and compute premetrics or geodesics, resulting in a simpler, faster, and fully intrinsic framework.

Clinical decisions are high-stakes and require explicit justification, making model interpretability essential for auditing deep clinical models prior to deployment. As the ecosystem of model architectures and explainability methods expands, critical questions remain: Do architectural features like attention improve explainability? Do interpretability approaches generalize across clinical tasks? While prior benchmarking efforts exist, they often lack extensibility and reproducibility, and critically, fail to systematically examine how interpretability varies across the interplay of clinical tasks and model architectures. To address these gaps, we present a comprehensive benchmark evaluating interpretability methods across diverse clinical prediction tasks and model architectures. Our analysis reveals that: (1) attention when leveraged properly is a highly efficient approach for faithfully interpreting model predictions; (2) black-box interpreters like KernelSHAP and LIME are computationally infeasible for time-series clinical prediction tasks; and (3) several interpretability approaches are too unreliable to be trustworthy. From our findings, we discuss several guidelines on improving interpretability within clinical predictive pipelines. To support reproducibility and extensibility, we provide our implementations via PyHealth, a well-documented open-source framework: https://github.com/sunlabuiuc/PyHealth.

Synthetic data generation through document rewriting has emerged as a promising technique for improving language model pretraining, yet most studies focus on English and do not systematically control for the quality of the source data being rewritten. We present a controlled study of how synthetic rewriting interacts with source data quality in the context of Portuguese continued pretraining. Starting from ClassiCC-PT, a Portuguese corpus annotated with STEM and Educational quality scores, we construct two 10B-token subsets at different quality levels and rewrite each into four styles using a 7B instruction-tuned model, producing approximately 40B tokens of synthetic data per condition. We train two English-centric base models (1.1B and 7B parameters) on each condition and evaluate on PoETa V2, a comprehensive 44-task Portuguese benchmark. At the 7B scale, rewriting high-quality data yields a +3.4 NPM gain over the same data unmodified, while rewriting low-quality data provides only +0.5 NPM. At the 1.1B scale, this interaction is weaker, with unmodified low-quality data performing comparably to rewritten high-quality data. Our results demonstrate that synthetic rewriting acts primarily as a quality multiplier rather than a substitute for data curation, and that this effect is scale-dependent.

Patch reviewing is critical for software development, especially in distributed open-source development, which highly depends on voluntary work, such as Linux. This paper studies the past 10 years of patch reviews of the Linux memory management subsystem to characterize the challenges involved in patch reviewing at scale. Our study reveals that the review process is still primarily reliant on human effort despite a wide-range of automatic checking tools. Although kernel developers strive to review all patch proposals, they struggle to keep up with the increasing volume of submissions and depend significantly on a few developers for these reviews. To help scale the patch review process, we introduce FLINT, a patch validation system framework that synthesizes insights from past discussions among developers and automatically analyzes patch proposals for compliance. FLINT employs a rule-based analysis informed by past discussions among developers and an LLM that does not require training or fine-tuning on new data, and can continuously improve with minimum human effort. FLINT uses a multi-stage approach to efficiently distill the essential information from past discussions. Later, when a patch proposal needs review, FLINT retrieves the relevant validation rules for validation and generates a reference-backed report that developers can easily interpret and validate. FLINT targets bugs that traditional tools find hard to detect, ranging from maintainability issues, e.g., design choices and naming conventions, to complex concurrency issues, e.g., deadlocks and data races. FLINT detected 2 new issues in Linux v6.18 development cycle and 7 issues in previous versions. FLINT achieves 21% and 14% of higher ground-truth coverage on concurrency bugs than the baseline with LLM only. Moreover, FLINT achieves a 35% false positive rate, which is lower than the baseline.

State-of-the-art crowd counting and localization are primarily modeled using two paradigms: density maps and point regression. Given the field's security ramifications, there is active interest in model robustness against adversarial attacks. Recent studies have demonstrated transferability across density-map-based approaches via adversarial patches, but cross-paradigm attacks (i.e., across both density map-based models and point regression-based models) remain unexplored. We introduce a novel adversarial framework that compromises both density map and point regression architectural paradigms through a comprehensive multi-task loss optimization. For point-regression models, we employ scene-density-specific high-confidence logit suppression; for density-map approaches, we use peak-targeted density map suppression. Both are combined with model-agnostic perceptual constraints to ensure that perturbations are effective and imperceptible to the human eye. Extensive experiments demonstrate the effectiveness of our attack, achieving on average a 7X increase in Mean Absolute Error compared to clean images while maintaining competitive visual quality, and successfully transferring across seven state-of-the-art crowd models with transfer ratios ranging from 0.55 to 1.69. Our approach strikes a balance between attack effectiveness and imperceptibility compared to state-of-the-art transferable attack strategies. The source code is available at https://github.com/simurgh7/CrowdGen

Diffusion models are increasingly used for robot learning, but current designs face a clear trade-off. Action-chunking diffusion policies like ManiCM are fast to run, yet they only predict short segments of motion. This makes them reactive, but unable to capture time-dependent motion primitives, such as following a spring-damper-like behavior with built-in dynamic profiles of acceleration and deceleration. Recently, Movement Primitive Diffusion (MPD) partially addresses this limitation by parameterizing full trajectories using Probabilistic Dynamic Movement Primitives (ProDMPs), thereby enabling the generation of temporally structured motions. Nevertheless, MPD integrates the motion decoder directly into a multi-step diffusion process, resulting in prohibitively high inference latency that limits its applicability in real-time control settings. We propose FODMP (Fast One-step Diffusion of Movement Primitives), a new framework that distills diffusion models into the ProDMPs trajectory parameter space and generates motion using a single-step decoder. FODMP retains the temporal structure of movement primitives while eliminating the inference bottleneck through single-step consistency distillation. This enables robots to execute time-dependent primitives at high inference speed, suitable for closed-loop vision-based control. On standard manipulation benchmarks (MetaWorld, ManiSkill), FODMP runs up to 10 times faster than MPD and 7 times faster than action-chunking diffusion policies, while matching or exceeding their success rates. Beyond speed, by generating fast acceleration-deceleration motion primitives, FODMP allows the robot to intercept and securely catch a fast-flying ball, whereas action-chunking diffusion policy and MPD respond too slowly for real-time interception.

Until recently, the success of large-scale vision-language models (VLMs) has primarily relied on billion-sample datasets, posing a significant barrier to progress. Latest works have begun to close this gap by improving supervision quality, but each addresses only a subset of the weaknesses in contrastive pretraining. We present GoldiCLIP, a framework built on a Goldilocks principle of finding the right balance of supervision signals. Our multifaceted training framework synergistically combines three key innovations: (1) a text-conditioned self-distillation method to align both text-agnostic and text-conditioned features; (2) an encoder integrated decoder with Visual Question Answering (VQA) objective that enables the encoder to generalize beyond the caption-like queries; and (3) an uncertainty-based weighting mechanism that automatically balances all heterogeneous losses. Trained on just 30 million images, 300x less data than leading methods, GoldiCLIP achieves state-of-the-art among data-efficient approaches, improving over the best comparable baseline by 2.2 points on MSCOCO retrieval, 2.0 on fine-grained retrieval, and 5.9 on question-based retrieval, while remaining competitive with billion-scale models. Project page: https://petsi.uk/goldiclip.

Computed tomography image segmentation of complex abdominal aortic aneurysms (AAA) often fails because the models assign internal focus to irrelevant structures or do not focus on thin, low-contrast targets. Where the model looks is the primary training signal, and thus we propose an Explainable AI (XAI) guided encoder shaping framework. Our method computes a dense, attribution-based encoder focus map ("XAI field") from the final encoder block and uses it in two complementary ways: (i) we align the predicted probability mass to the XAI field to promote agreement between focus and output; and (ii) we route the field into a lightweight refinement pathway and a confidence prior that modulates logits at inference, suppressing distractors while preserving subtle structures. The objective terms serve only as control signals; the contribution is the integration of attribution guidance into representation and decoding. We evaluate clinically validated challenging cases curated for failure-prone scenarios. Compared to a base SAM setup, our implementation yields substantial improvements. The observed gains suggest that explicitly optimizing encoder focus via XAI guidance is a practical and effective principle for reliable segmentation in complex scenarios.

To fully capture the meaning of a sentence, semantic representations should encode aspect, which describes the internal temporal structure of events. In graph-based meaning representation frameworks such as Uniform Meaning Representations (UMR), aspect lets one know how events unfold over time, including distinctions such as states, activities, and completed events. Despite its importance, aspect remains sparsely annotated across semantic meaning representation frameworks. This has, in turn, hindered not only current manual annotation, but also the development of automatic systems capable of predicting aspectual information. In this paper, we introduce a new dataset of English sentences annotated with UMR aspect labels over Abstract Meaning Representation (AMR) graphs that lack the feature. We describe the annotation scheme and guidelines used to label eventive predicates according to the UMR aspect lattice, as well as the annotation pipeline used to ensure consistency and quality across annotators through a multi-step adjudication process. To demonstrate the utility of our dataset for future automation, we present baseline experiments using three modeling approaches. Our results establish initial benchmarks for automatic UMR aspect prediction and provide a foundation for integrating aspect into semantic meaning representations more broadly.

While backpropagation and automatic differentiation have driven deep learning's success, the physical limits of chip manufacturing and rising environmental costs of deep learning motivate alternative learning paradigms such as physical neural networks. However, most existing physical neural networks still rely on digital computing for training, largely because backpropagation and automatic differentiation are difficult to realize in physical systems. We introduce FFzero, a forward-only learning framework enabling stable neural network training without backpropagation or automatic differentiation. FFzero combines layer-wise local learning, prototype-based representations, and directional-derivative-based optimization through forward evaluations only. We show that local learning is effective under forward-only optimization, where backpropagation fails. FFzero generalizes to multilayer perceptron and convolutional neural networks across classification and regression. Using a simulated photonic neural network as an example, we demonstrate that FFzero provides a viable path toward backpropagation-free in-situ physical learning.

Routing has emerged as a promising strategy for balancing performance and cost in large language model (LLM) systems that combine lightweight models with powerful but expensive large models. Recent studies show that \emph{probe routing}, which predicts the correctness of a small model using its hidden states, provides an effective solution in text-only LLMs. However, we observe that these probes degrade substantially when applied to multimodal LLMs (MLLMs). Through empirical analysis, we find that the presence of visual inputs weakens the separability of correctness signals in hidden states, making them harder to extract using standard probe designs. To address this challenge, we introduce two complementary approaches for improving probe routing in MLLMs. First, we propose the \emph{Attention Probe}, which aggregates hidden states from the preceding layer based on attention scores to recover distributed correctness signals. Second, we present the \emph{KL-Regularized LoRA Probe (ReLope)}, which inserts a lightweight LoRA adapter and applies a KL regularizer to learn routing-aware representations. Comprehensive experiments show that our methods consistently outperform baselines, suggesting that improving the quality of hidden states is key to effective routing in MLLMs. Our code is available at https://github.com/Spinozaaa/ReLope.

Effective problem solving with Large Language Models (LLMs) can be enhanced when they are paired with external search algorithms. By viewing the space of diverse ideas and their follow-up possibilities as a tree structure, the search algorithm can navigate such a search space and guide the LLM toward better solutions more efficiently. While the search algorithm enables an effective balance between exploitation and exploration of a tree-structured space, the need for an external component can complicate the overall problem-solving process. We therefore pose the following question: Can LLMs or their underlying Transformer architectures approximate a search algorithm? To answer this question, we first introduce a simplified framework in which tree extensions and feedback signals are externally specified, allowing for controlled evaluation of search capabilities. We call this setting unknown tree search with bandit feedback. Within this setting, we show that Transformers are theoretically expressive enough to implement distinct search strategies and can be trained from scratch to approximate those strategies. Our Transformer models exhibit the possibility of generalizing to unseen conditions such as longer horizons or deeper trees. Furthermore, we demonstrate that continued task-focused training unlocks the complete capabilities of a pretrained LLM, by fine-tuning the LLM on search trajectories.

AI agents increasingly call tools via the Model Context Protocol (MCP) and delegate to other agents via Agent-to-Agent (A2A), yet neither protocol verifies agent identity. A scan of approximately 2,000 MCP servers found all lacked authentication. In our survey, we did not identify a prior implemented protocol that jointly combines public-key verifiable delegation, holder-side attenuation, expressive chained policy, transport bindings across MCP/A2A/HTTP, and provenance-oriented completion records. We introduce Invocation-Bound Capability Tokens (IBCTs), a primitive that fuses identity, attenuated authorization, and provenance binding into a single append-only token chain. IBCTs operate in two wire formats: compact mode (a signed JWT for single-hop cases) and chained mode (a Biscuit token with Datalog policies for multi-hop delegation). We provide reference implementations in Python and Rust with full cross-language interoperability. Compact mode verification takes 0.049ms (Rust) and 0.189ms (Python), with 0.22ms overhead over no-auth in real MCP-over-HTTP deployment. In a real multi-agent deployment with Gemini 2.5 Flash, AIP adds 2.35ms of overhead (0.086% of total end-to-end latency). Adversarial evaluation across 600 attack attempts shows 100% rejection rate, with two attack categories (delegation depth violation and audit evasion through empty context) uniquely caught by AIP's chained delegation model that neither unsigned nor plain JWT deployments detect.

This article discusses the challenges of testing software systems with increasingly integrated AI and LLM functionalities. LLMs are powerful but unreliable, and labeled ground truth for testing rarely scales. Metamorphic Testing solves this by turning relations among multiple test executions into executable test oracles.

Milestone Driven Agile Execution is a hybrid management framework where the empirical control component of agile development is retained but the prioritization of the backlog is done according to a macro or strategic (milestone) plan that drives the execution of the project. MDAX is method agnostic, in the sense that the development approach is not embedded in the execution mechanism but in the plan that drives it. This allows organizations using it to choose the development approach that suites them most,

Clinical documentation is a critical factor for patient safety, diagnosis, and continuity of care. The administrative burden of EHRs is a significant factor in physician burnout. This is a critical issue for low-resource languages, including Finnish. This study aims to investigate the effectiveness of a domain-aligned natural language processing (NLP); large language model for medical transcription in Finnish by fine-tuning LLaMA 3.1-8B on a small validated corpus of simulated clinical conversations by students at Metropolia University of Applied Sciences. The fine-tuning process for medical transcription used a controlled preprocessing and optimization approach. The fine-tuning effectiveness was evaluated by sevenfold cross-validation. The evaluation metrics for fine-tuned LLaMA 3.1-8B were BLEU = 0.1214, ROUGE-L = 0.4982, and BERTScore F1 = 0.8230. The results showed a low n-gram overlap but a strong semantic similarity with reference transcripts. This study indicate that fine-tuning can be an effective approach for translation of medical discourse in spoken Finnish and support the feasibility of fine-tuning a privacy-oriented domain-specific large language model for clinical documentation in Finnish. Beside that provide directions for future work.

Systematic reviews traditionally have taken considerable amounts of human time and energy to complete, in part due to the extensive number of titles and abstracts that must be reviewed for potential inclusion. Recently, researchers have begun to explore how to use large language models (LLMs) to make this process more efficient. However, research to date has shown inconsistent results. We posit this is because prompting alone may not provide sufficient context for the model(s) to perform well. In this study, we fine-tune a small 1.2 billion parameter open-weight LLM specifically for study screening in the context of a systematic review in which humans rated more than 8500 titles and abstracts for potential inclusion. Our results showed strong performance improvements from the fine-tuned model, with the weighted F1 score improving 80.79% compared to the base model. When run on the full dataset of 8,277 studies, the fine-tuned model had 86.40% agreement with the human coder, a 91.18% true positive rate, a 86.38% true negative rate, and perfect agreement across multiple inference runs. Taken together, our results show that there is promise for fine-tuning LLMs for title and abstract screening in large-scale systematic reviews.

The engineering design research community has studied agentic AI systems that use Large Language Model (LLM) agents to automate the engineering design process. However, these systems are prone to some of the same pathologies that plague humans. Just as human designers, LLM design agents can fixate on existing paradigms and fail to explore alternatives when solving design challenges, potentially leading to suboptimal solutions. In this work, we propose (1) a novel Self-Regulation Loop (SRL), in which the Design Agent self-regulates and explicitly monitors its own metacognition, and (2) a novel Co-Regulation Design Agentic Loop (CRDAL), in which a Metacognitive Co-Regulation Agent assists the Design Agent in metacognition to mitigate design fixation, thereby improving system performance for engineering design tasks. In the battery pack design problem examined here, we found that the novel CRDAL system generates designs with better performance, without significantly increasing the computational cost, compared to a plain Ralph Wiggum Loop (RWL) and the metacognitively self-assessing Self-Regulation Loop (SRL). Also, we found that the CRDAL system navigated through the latent design space more effectively than both SRL and RWL. However, the SRL did not generate designs with significantly better performance than RWL, even though it explored a different region of the design space. The proposed system architectures and findings of this work provide practical implications for future development of agentic AI systems for engineering design.

Synthetic cardiac MRI (CMRI) generation has emerged as a promising strategy to overcome the scarcity of annotated medical imaging data. Recent advances in GANs, VAEs, diffusion probabilistic models, and flow-matching techniques aim to generate anatomically accurate images while addressing challenges such as limited labeled datasets, vendor variability, and risks of privacy leakage through model memorization. Maskconditioned generation improves structural fidelity by guiding synthesis with segmentation maps, while diffusion and flowmatching models offer strong boundary preservation and efficient deterministic transformations. Cross-domain generalization is further supported through vendor-style conditioning and preprocessing steps like intensity normalization. To ensure privacy, studies increasingly incorporate membership inference attacks, nearest-neighbor analyses, and differential privacy mechanisms. Utility evaluations commonly measure downstream segmentation performance, with evidence showing that anatomically constrained synthetic data can enhance accuracy and robustness across multi-vendor settings. This review aims to compare existing CMRI generation approaches through the lenses of fidelity, utility, and privacy, highlighting current limitations and the need for integrated, evaluation-driven frameworks for reliable clinical workflows.

Conditional independence is central to modern statistics, but beyond special parametric families it rarely admits an exact covariance characterization. We introduce the binary expansion group intersection network (BEGIN), a distribution-free graphical representation for multivariate binary data and bit-encoded multinomial variables. For arbitrary binary random vectors and bit representations of multinomial variables, we prove that conditional independence is equivalent to a sparse linear representation of conditional expectations, to a block factorization of the corresponding interaction covariance matrix, and to block diagonality of an associated generalized Schur complement. The resulting graph is indexed by the intersection of multiplicative groups of binary interactions, yielding an analogue of Gaussian graphical modeling beyond the Gaussian setting. This viewpoint treats data bits as atoms and local BEGIN molecules as building blocks for large Markov random fields. We also show how dyadic bit representations allow BEGIN to approximate conditional independence for general random vectors under mild regularity conditions. A key technical device is the Hadamard prism, a linear map that links interaction covariances to group structure.

Software development is iterative, yet agentic coding benchmarks overwhelmingly evaluate single-shot solutions against complete specifications. Code can pass the test suite but become progressively harder to extend. Recent iterative benchmarks attempt to close this gap, but constrain the agent's design decisions too tightly to faithfully measure how code quality shapes future extensions. We introduce SlopCodeBench, a language-agnostic benchmark comprising 20 problems and 93 checkpoints, in which agents repeatedly extend their own prior solutions under evolving specifications that force architectural decisions without prescribing internal structure. We track two trajectory-level quality signals: verbosity, the fraction of redundant or duplicated code, and structural erosion, the share of complexity mass concentrated in high-complexity functions. No agent solves any problem end-to-end across 11 models; the highest checkpoint solve rate is 17.2%. Quality degrades steadily: erosion rises in 80% of trajectories and verbosity in 89.8%. Against 48 open-source Python repositories, agent code is 2.2x more verbose and markedly more eroded. Tracking 20 of those repositories over time shows that human code stays flat, while agent code deteriorates with each iteration. A prompt-intervention study shows that initial quality can be improved, but it does not halt degradation. These results demonstrate that pass-rate benchmarks systematically undermeasure extension robustness, and that current agents lack the design discipline iterative software development demands.

Standard vision models treat objects as independent points in Euclidean space, unable to capture hierarchical structure like parts within wholes. We introduce Worldline Slot Attention, which models objects as persistent trajectories through spacetime worldlines, where each object has multiple slots at different hierarchy levels sharing the same spatial position but differing in temporal coordinates. This architecture consistently fails without geometric structure: Euclidean worldlines achieve 0.078 level accuracy, below random chance (0.33), while Lorentzian worldlines achieve 0.479-0.661 across three datasets: a 6x improvement replicated over 20+ independent runs. Lorentzian geometry also outperforms hyperbolic embeddings showing visual hierarchies require causal structure (temporal dependency) rather than tree structure (radial branching). Our results demonstrate that hierarchical object discovery requires geometric structure encoding asymmetric causality, an inductive bias absent from Euclidean space but natural to Lorentzian light cones, achieved with only 11K parameters. The code is available at: https://github.com/iclrsubmissiongram/loco.

Machine-learned interatomic potentials (MLIPs), particularly graph neural network (GNN)-based models, offer a promising route to achieving near-density functional theory (DFT) accuracy at significantly reduced computational cost. However, their practical deployment is often limited by the large volumes of expensive quantum mechanical training data required. In this work, we introduce a transfer learning framework, Transfer-PaiNN (T-PaiNN), that substantially improves the data efficiency of GNN-MLIPs by leveraging inexpensive classical force field data. The approach consists of pretraining a PaiNN MLIP architecture on large-scale datasets generated from classical molecular simulations, followed by fine-tuning (dubbed autotuning) using a comparatively small DFT dataset. We demonstrate the effectiveness of autotuning T-PaiNN on both gas-phase molecular systems (QM9 dataset) and condensed-phase liquid water. Across all cases, T-PaiNN significantly outperforms models trained solely on DFT data, achieving order-of-magnitude reductions in mean absolute error while accelerating training convergence. For example, using the QM9 data set, error reductions of up to 25 times are observed in low-data regimes, while liquid water simulations show improved predictions of energies, forces, and experimentally relevant properties such as density and diffusion. These gains arise from the model's ability to learn general features of the potential energy surface from extensive classical sampling, which are subsequently refined to quantum accuracy. Overall, this work establishes transfer learning from classical force fields as a practical and computationally efficient strategy for developing high-accuracy, data-efficient GNN interatomic potentials, enabling broader application of MLIPs to complex chemical systems.

Online Health Communities connect patients for peer support, but users face a discovery challenge when they have minimal prior interactions to guide personalization. We study recommendation under extreme interaction sparsity in a survey driven setting where each user provides a 16 dimensional intake vector and each support group has a structured feature profile. We extend Neural Collaborative Filtering architectures, including Matrix Factorization, Multi Layer Perceptron, and NeuMF, with an auxiliary pseudo label objective derived from survey group feature alignment using cosine similarity mapped to [0, 1]. The resulting Pseudo Label NCF learns dual embedding spaces: main embeddings for ranking and pseudo label embeddings for semantic alignment. We evaluate on a dataset of 165 users and 498 support groups using a leave one out protocol that reflects cold start conditions. All pseudo label variants improve ranking performance: MLP improves HR@5 from 2.65% to 5.30%, NeuMF from 4.46% to 5.18%, and MF from 4.58% to 5.42%. Pseudo label embedding spaces also show higher cosine silhouette scores than baseline embeddings, with MF improving from 0.0394 to 0.0684 and NeuMF from 0.0263 to 0.0653. We further observe a negative correlation between embedding separability and ranking accuracy, indicating a trade off between interpretability and performance. These results show that survey derived pseudo labels improve recommendation under extreme sparsity while producing interpretable task specific embedding spaces.

The emergence of large language model agents capable of invoking external tools has created urgent need for formal verification of agent protocols. Two paradigms dominate this space: Schema-Guided Dialogue (SGD), a research framework for zero-shot API generalization, and the Model Context Protocol (MCP), an industry standard for agent-tool integration. While both enable dynamic service discovery through schema descriptions, their formal relationship remains unexplored. Building on prior work establishing the conceptual convergence of these paradigms, we present the first process calculus formalization of SGD and MCP, proving they are structurally bisimilar under a well-defined mapping Phi. However, we demonstrate that the reverse mapping Phi^{-1} is partial and lossy, revealing critical gaps in MCP's expressivity. Through bidirectional analysis, we identify five principles -- semantic completeness, explicit action boundaries, failure mode documentation, progressive disclosure compatibility, and inter-tool relationship declaration -- as necessary and sufficient conditions for full behavioral equivalence. We formalize these principles as type-system extensions MCP+, proving MCP+ is isomorphic to SGD. Our work provides the first formal foundation for verified agent systems and establishes schema quality as a provable safety property.

Grokking -- the delayed onset of generalization after early memorization -- is often described with phase-transition language, but that claim has lacked falsifiable finite-size inputs. Here we supply those inputs by treating the group order $p$ of $\mathbb{Z}_p$ as an admissible extensive variable and a held-out spectral head-tail contrast as a representation-level order parameter, then apply a condensed-matter-style diagnostic chain to coarse-grid sweeps and a dense near-critical addition audit. Binder-like crossings reveal a shared finite-size boundary, and susceptibility comparison strongly disfavors a smooth-crossover interpretation ($Δ\mathrm{AIC}=16.8$ in the near-critical audit). Phase-transition language in grokking can therefore be tested as a quantitative finite-size claim rather than invoked as analogy alone, although the transition order remains unresolved at present.

Weather data, comprising multiple variables, poses significant challenges due to its high dimensionality and multimodal nature. Creating low-dimensional embeddings requires compressing this data into a compact, shared latent space. This compression is required to improve the efficiency and performance of downstream tasks, such as forecasting or extreme-weather detection. Self-supervised learning, particularly contrastive learning, offers a way to generate low-dimensional, robust embeddings from unlabelled data, enabling downstream tasks when labelled data is scarce. Despite initial exploration of contrastive learning in weather data, particularly with the ERA5 dataset, the current literature does not extensively examine its benefits relative to alternative compression methods, notably autoencoders. Moreover, current work on contrastive learning does not investigate how these models can incorporate sparse data, which is more common in real-world data collection. It is critical to explore and understand how contrastive learning contributes to creating more robust embeddings for sparse weather data, thereby improving performance on downstream tasks. Our work extensively explores contrastive learning on the ERA5 dataset, aligning sparse samples with complete ones via a contrastive loss term to create SPARse-data augmented conTRAstive spatiotemporal embeddings (SPARTA). We introduce a temporally aware batch sampling strategy and a cycle-consistency loss to improve the structure of the latent space. Furthermore, we propose a novel graph neural network fusion technique to inject domain-specific physical knowledge. Ultimately, our results demonstrate that contrastive learning is a feasible and advantageous compression method for sparse geoscience data, thereby enhancing performance in downstream tasks.

AI safety is an increasingly urgent concern as the capabilities and adoption of AI systems grow. Existing evolutionary models of AI governance have primarily examined incentives for safe development and effective regulation, typically representing users' trust as a one-shot adoption choice rather than as a dynamic, evolving process shaped by repeated interactions. We instead model trust as reduced monitoring in a repeated, asymmetric interaction between users and AI developers, where checking AI behaviour is costly. Using evolutionary game theory, we study how user trust strategies and developer choices between safe (compliant) and unsafe (non-compliant) AI co-evolve under different levels of monitoring cost and institutional regimes. We complement the infinite-population replicator analysis with stochastic finite-population dynamics and reinforcement learning (Q-learning) simulations. Across these approaches, we find three robust long-run regimes: no adoption with unsafe development, unsafe but widely adopted systems, and safe systems that are widely adopted. Only the last is desirable, and it arises when penalties for unsafe behaviour exceed the extra cost of safety and users can still afford to monitor at least occasionally. Our results formally support governance proposals that emphasise transparency, low-cost monitoring, and meaningful sanctions, and they show that neither regulation alone nor blind user trust is sufficient to prevent evolutionary drift towards unsafe or low-adoption outcomes.

Efficient task scheduling in large-scale distributed systems presents significant challenges due to dynamic workloads, heterogeneous resources, and competing quality-of-service requirements. Traditional centralized approaches face scalability limitations and single points of failure, while classical heuristics lack adaptability to changing conditions. This paper proposes a decentralized multi-agent deep reinforcement learning (DRL-MADRL) framework for task scheduling in heterogeneous distributed systems. We formulate the problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) and develop a lightweight actor-critic architecture implemented using only NumPy, enabling deployment on resource-constrained edge devices without heavyweight machine learning frameworks. Using workload characteristics derived from the publicly available Google Cluster Trace dataset, we evaluate our approach on a 100-node heterogeneous system processing 1,000 tasks per episode over 30 experimental runs. Experimental results demonstrate 15.6% improvement in average task completion time (30.8s vs 36.5s for random baseline), 15.2% energy efficiency gain (745.2 kWh vs 878.3 kWh), and 82.3% SLA satisfaction compared to 75.5% for baselines, with all improvements statistically significant (p < 0.001). The lightweight implementation requires only NumPy, Matplotlib, and SciPy. Complete source code and experimental data are provided for full reproducibility at https://github.com/danielbenniah/marl-distributed-scheduling.

In the design and safety analysis of advanced reactor systems, constructing input files for system-level thermal-hydraulics codes such as the System Analysis Module (SAM) remains a labor-intensive task. Analysts must extract and reconcile design data from heterogeneous engineering documents and manually translate it into solver-specific syntax. In this paper, we present AutoSAM, an agentic framework that automates SAM input file generation. The framework combines a large language model agent with retrieval-augmented generation over the solver's user guide and theory manual, together with specialized tools for analyzing PDFs, images, spreadsheets, and text files. AutoSAM ingests unstructured engineering documents, including system diagrams, design reports, and data tables, extracts simulation-relevant parameters into a human-auditable intermediate representation, and synthesizes validated, solver-compatible input decks. Its multimodal retrieval pipeline integrates scientific text extraction, vision-based figure interpretation, semantic embedding, and query answering. We evaluate AutoSAM on four case studies of increasing complexity: a single-pipe steady-state model, a solid-fuel channel with temperature reactivity feedback, the Advanced Burner Test Reactor core, and the Molten Salt Reactor Experiment primary loop. Across all cases, the agent produces runnable SAM models consistent with expected thermal-hydraulic behavior while explicitly identifying missing data and labeling assumed values. The framework achieves 100% utilization of structured inputs, about 88% extraction from PDF text, and 100% completeness in vision-based geometric extraction. These results demonstrate a practical path toward prompt-driven reactor modeling, in which analysts provide system descriptions and supporting documentation while the agent translates them into transparent, and executable, SAM simulations.

Unified Modeling Language (UML) is widely used for modeling IT systems but lacks formal rules to ensure consistency across diagrams. This often leads to inconsistencies when shared elements are interpreted differently. To address this, architects use consistency rules that derive elements in target diagrams from more abstract source diagrams. However, these rules are often written in natural language and applied at the element level, making them difficult to reuse or integrate with modeling tools. This paper introduces composite consistency rules-higher-level patterns that combine simple rules into more intuitive, reusable structures. These rules reflect architects design practices and support systematic, error-resistant model development. Implemented as JScript scripts in Sparx Enterprise Architect, they improve automation, reduce redundancy, and accelerate design. Composite rules enhance the consistency and completeness of UML architectures and can be reused across projects. They also support pattern-driven modeling and open possibilities for AI-assisted architecture generation and code integration.

Appearance-based gaze estimation frequently relies on deep Convolutional Neural Networks (CNNs). These models are accurate, but computationally expensive and act as "black boxes", offering little interpretability. Geometric methods based on facial landmarks are a lightweight alternative, but their performance limits and generalization capabilities remain underexplored in modern benchmarks. In this study, we conduct a comprehensive evaluation of landmark-based gaze estimation. We introduce a standardized pipeline to extract and normalize landmarks from three large-scale datasets (Gaze360, ETH-XGaze, and GazeGene) and train lightweight regression models, specifically Extreme Gradient Boosted trees and two neural architectures: a holistic Multi-Layer Perceptron (MLP) and a siamese MLP designed to capture binocular geometry. We find that landmark-based models exhibit lower performance in within-domain evaluation, likely due to noise introduced into the datasets by the landmark detector. Nevertheless, in cross-domain evaluation, the proposed MLP architectures show generalization capabilities comparable to those of ResNet18 baselines. These findings suggest that sparse geometric features encode sufficient information for robust gaze estimation, paving the way for efficient, interpretable, and privacy-friendly edge applications. The source code and generated landmark-based datasets are available at https://github.com/daniele-agostinelli/LandmarkGaze.git.

Spatial reasoning focuses on locating target objects based on spatial relations in 3D scenes, which plays a crucial role in developing intelligent embodied agents. Due to the limited availability of 3D scene-language paired data, it is challenging to train models with strong reasoning ability from scratch. Previous approaches have attempted to inject 3D scene representations into the input space of Large Language Models (LLMs) and leverage the pretrained comprehension and reasoning abilities for spatial reasoning. However, models encoding absolute positions struggle to extract spatial relations from prematurely fused features, while methods explicitly encoding all spatial relations (which is quadratic in the number of objects) as input tokens suffer from poor scalability. To address these limitations, we propose QuatRoPE, a novel positional embedding method with an input length that is linear to the number of objects, and explicitly calculates pairwise spatial relations through the dot product in attention layers. QuatRoPE's holistic vector encoding of 3D coordinates guarantees a high degree of spatial consistency, maintaining fidelity to the scene's geometric integrity. Additionally, we introduce the Isolated Gated RoPE Extension (IGRE), which effectively limits QuatRoPE's influence to object-related tokens, thereby minimizing interference with the LLM's existing positional embeddings and maintaining the LLM's original capabilities. Extensive experiments demonstrate the effectiveness of our approaches. The code and data are available at https://github.com/oceanflowlab/QuatRoPE.

Analog design often slows down because even small changes to device sizes or biases require expensive simulation cycles, and high-quality solutions typically occupy only a narrow part of a very large search space. While existing optimizers reduce some of this burden, they largely operate without the kind of judgment designers use when deciding where to search next. This paper presents an actor-critic optimization framework (ACOF) for analog sizing that brings that form of guidance into the loop. Rather than treating optimization as a purely black-box search problem, ACOF separates the roles of proposal and evaluation: an actor suggests promising regions of the design space, while a critic reviews those choices, enforces design legality, and redirects the search when progress is hampered. This structure preserves compatibility with standard simulator-based flows while making the search process more deliberate, stable, and interpretable. Across our test circuits, ACOF improves the top-10 figure of merit by an average of 38.9% over the strongest competing baseline and reduces regret by an average of 24.7%, with peak gains of 70.5% in FoM and 42.2% lower regret on individual circuits. By combining iterative reasoning with simulation-driven search, the framework offers a more transparent path toward automated analog sizing across challenging design spaces.

Multi-step tool orchestration, where LLMs must invoke multiple dependent APIs in the correct order while propagating intermediate outputs, remains challenging. State-of-the-art models frequently fail on full sequence execution, with parameter value errors accounting for a significant portion of failures. Training models to handle such workflows faces two obstacles: existing environments focus on simple per-turn function calls with simulated data, and binary rewards provide no signal for partial correctness. We present a framework addressing both challenges. First, we construct a reinforcement learning environment backed by a large-scale cache of real API responses, enabling a data synthesis pipeline that samples valid multi-step orchestration traces with controllable complexity and significantly higher generation efficiency than unconstrained methods. Second, we propose a graduated reward design that decomposes correctness into atomic validity (individual function call correctness at increasing granularity) and orchestration (correct tool sequencing with dependency respect). On ComplexFuncBench, our approach demonstrates substantial improvements in turn accuracy. Ablation studies confirm both reward components are essential: using either alone significantly degrades performance.

Discrete choice models are fundamental tools in management science, economics, and marketing for understanding and predicting decision-making. Logit-based models are dominant in applied work, largely due to their convenient closed-form expressions for choice probabilities. However, these models entail restrictive assumptions on the stochastic utility component, constraining our ability to capture realistic and theoretically grounded choice behavior$-$most notably, substitution patterns. In this work, we propose an amortized inference approach using a neural network emulator to approximate choice probabilities for general error distributions, including those with correlated errors. Our proposal includes a specialized neural network architecture and accompanying training procedures designed to respect the invariance properties of discrete choice models. We provide group-theoretic foundations for the architecture, including a proof of universal approximation given a minimal set of invariant features. Once trained, the emulator enables rapid likelihood evaluation and gradient computation. We use Sobolev training, augmenting the likelihood loss with a gradient-matching penalty so that the emulator learns both choice probabilities and their derivatives. We show that emulator-based maximum likelihood estimators are consistent and asymptotically normal under mild approximation conditions, and we provide sandwich standard errors that remain valid even with imperfect likelihood approximation. Simulations show significant gains over the GHK simulator in accuracy and speed.

In deploying artificial intelligence (AI) models, selective prediction offers the option to abstain from making a prediction when uncertain about model quality. To fulfill its promise, it is crucial to enforce strict and precise error control over cases where the model is trusted. We propose Selective Conformal Risk control with E-values (SCoRE), a new framework for deriving such decisions for any trained model and any user-defined, bounded and continuously-valued risk. SCoRE offers two types of guarantees on the risk among ``positive'' cases in which the system opts to trust the model. Built upon conformal inference and hypothesis testing ideas, SCoRE first constructs a class of (generalized) e-values, which are non-negative random variables whose product with the unknown risk has expectation no greater than one. Such a property is ensured by data exchangeability without requiring any modeling assumptions. Passing these e-values on to hypothesis testing procedures, we yield the binary trust decisions with finite-sample error control. SCoRE avoids the need of uniform concentration, and can be readily extended to settings with distribution shifts. We evaluate the proposed methods with simulations and demonstrate their efficacy through applications to error management in drug discovery, health risk prediction, and large language models.

AI assistants can decompose multi-step workflows, but they do not natively speak industrial protocols such as Modbus, MQTT/Sparkplug B, or OPC UA, so this paper presents INDUSTRICONNECT, a prototype suite of Model Context Protocol (MCP) adapters that expose industrial operations as schema-discoverable AI tools while preserving protocol-specific connectivity and safety controls; the system uses a common response envelope and a mock-first workflow so adapter behavior can be exercised locally before connecting to plant equipment, and a deterministic benchmark covering normal, fault-injected, stress, and recovery scenarios evaluates the flagship adapters, comprising 870 runs (480 normal, 210 fault-injected, 120 stress, 60 recovery trials) and 2820 tool calls across 7 fault scenarios and 12 stress scenarios, where the normal suite achieved full success, the fault suite confirmed structured error handling with adapter-level uint16 range validation, the stress suite identified concurrency boundaries, and same-session recovery after endpoint restart is demonstrated for all three protocols, with results providing evidence spanning adapter correctness, concurrency behavior, and structured error handling for AI-assisted industrial operations.

Random cropping is one of the most common data augmentation techniques in computer vision, yet the role of its inherent randomness in training differentially private machine learning models has thus far gone unexplored. We observe that when sensitive content in an image is spatially localized, such as a face or license plate, random cropping can probabilistically exclude that content from the model's input. This introduces a third source of stochasticity in differentially private training with stochastic gradient descent, in addition to gradient noise and minibatch sampling. This additional randomness amplifies differential privacy without requiring changes to model architecture or training procedure. We formalize this effect by introducing a patch-level neighboring relation for vision data and deriving tight privacy bounds for differentially private stochastic gradient descent (DP-SGD) when combined with random cropping. Our analysis quantifies the patch inclusion probability and shows how it composes with minibatch sampling to yield a lower effective sampling rate. Empirically, we validate that patch-level amplification improves the privacy-utility trade-off across multiple segmentation architectures and datasets. Our results demonstrate that aligning privacy accounting with domain structure and additional existing sources of randomness can yield stronger guarantees at no additional cost.

Spiking neural networks (SNNs) are promising for neuromorphic computing, but high-performing models still rely on dense multilayer architectures with substantial communication and state-storage costs. Inspired by autapses, we propose time-delayed autapse SNN (TDA-SNN), a framework that reconstructs SNNs with a single leaky integrate-and-fire neuron and a prototype-learning-based training strategy. By reorganizing internal temporal states, TDA-SNN can realize reservoir, multilayer perceptron, and convolution-like spiking architectures within a unified framework. Experiments on sequential, event-based, and image benchmarks show competitive performance in reservoir and MLP settings, while convolutional results reveal a clear space--time trade-off. Compared with standard SNNs, TDA-SNN greatly reduces neuron count and state memory while increasing per-neuron information capacity, at the cost of additional temporal latency in extreme single-neuron settings. These findings highlight the potential of temporally multiplexed single-neuron models as compact computational units for brain-inspired computing.

Multi-agent systems powered by large language models (LLMs) are increasingly deployed in settings that shape consequential decisions, both directly and indirectly. Yet it remains unclear whether their outcomes reflect collective reasoning, systematic bias, or mere chance. Recent work has sharpened this question with naming games, showing that even when no individual agent favors any label a priori, populations rapidly break symmetry and reach consensus. Here, we reveal the mechanism by introducing a minimal model, Quantized Simplex Gossip (QSG), and trace the microscopic origin of this agreement to mutual in-context learning. In QSG, agents maintain internal belief states but learn from one another's sampled outputs, so one agent's arbitrary choice becomes the next agent's evidence and can compound toward agreement. By analogy with neutral evolution, we call this sampling-driven regime memetic drift. QSG predicts a crossover from a drift-dominated regime, where consensus is effectively a lottery, to a selection regime, where weak biases are amplified and shape the outcome. We derive scaling laws for drift-induced polarization as a function of population size, communication bandwidth, in-context adaptation rate, and agents' internal uncertainty, and we validate them in both QSG simulations and naming-game experiments with LLM populations. Together, these results provide a framework for studying the collective mechanisms of social representation formation in multi-agent systems.

This article presents an argument for why quantum computers could unlock new methods for machine learning. We argue that spectral methods, in particular those that learn, regularise, or otherwise manipulate the Fourier spectrum of a machine learning model, are often natural for quantum computers. For example, if a generative machine learning model is represented by a quantum state, the Quantum Fourier Transform allows us to manipulate the Fourier spectrum of the state using the entire toolbox of quantum routines, an operation that is usually prohibitive for classical models. At the same time, spectral methods are surprisingly fundamental to machine learning: A spectral bias has recently been hypothesised to be the core principle behind the success of deep learning; support vector machines have been known for decades to regularise in Fourier space, and convolutional neural nets build filters in the Fourier space of images. Could, then, quantum computing open fundamentally different, much more direct and resource-efficient ways to design the spectral properties of a model? We discuss this potential in detail here, hoping to stimulate a direction in quantum machine learning research that puts the question of ``why quantum?'' first.

Network pruning, which removes less important parameters or architectures, is often expected to improve efficiency while preserving performance. However, this expectation does not consistently hold across language tasks: pruned models can perform well on non-generative tasks but frequently fail in generative settings. To understand this discrepancy, we analyze network pruning from a representation-hierarchy perspective, decomposing the internal computation of language models into three sequential spaces: embedding (hidden representations), logit (pre-softmax outputs), and probability (post-softmax distributions). We find that representations in the embedding and logit spaces are largely robust to pruning-induced perturbations. However, the nonlinear transformation from logits to probabilities amplifies these deviations, which accumulate across time steps and lead to substantial degradation during generation. In contrast, the stability of the categorical-token probability subspace, together with the robustness of the embedding space, supports the effectiveness of pruning for non-generative tasks such as retrieval and multiple-choice selection. Our analysis disentangles the effects of pruning across tasks and provides practical guidance for its application. Code is available at https://github.com/CASE-Lab-UMD/Pruning-on-Representations

Automatic depression detection from doctor-patient conversations has gained momentum thanks to the availability of public corpora and advances in language modeling. However, interpretability remains limited: strong performance is often reported without revealing what drives predictions. We analyze three datasets: ANDROIDS, DAIC-WOZ, E-DAIC and identify a systematic bias from interviewer prompts in semi-structured interviews. Models trained on interviewer turns exploit fixed prompts and positions to distinguish depressed from control subjects, often achieving high classification scores without using participant language. Restricting models to participant utterances distributes decision evidence more broadly and reflects genuine linguistic cues. While semi-structured protocols ensure consistency, including interviewer prompts inflates performance by leveraging script artifacts. Our results highlight a cross-dataset, architecture-agnostic bias and emphasize the need for analyses that localize decision evidence by time and speaker to ensure models learn from participants' language.

Anomaly detection is a core service in the Internet of Underwater Things, yet training accurate distributed models underwater is difficult because acoustic links are low-bandwidth, energy-intensive, and often unable to support direct sensor-to-surface communication. Standard flat federated learning therefore faces two coupled limitations in underwater deployments: expensive long-range transmissions and reduced participation when only a subset of sensors can reach the gateway. This paper proposes an energy-efficient hierarchical federated learning framework for underwater anomaly detection based on three components: feasibility-aware sensor-to-fog association, compressed model-update transmission, and selective cooperative aggregation among fog nodes. The proposed three-tier architecture localises most communication within short-range clusters while activating fog-to-fog exchange only when smaller clusters can benefit from nearby larger neighbours. A physics-grounded underwater acoustic model is used to evaluate detection quality, communication energy, and network participation jointly. In large synthetic deployments, only about 48% of sensors can directly reach the gateway in the 200-sensor case, whereas hierarchical learning preserves full participation through feasible fog paths. Selective cooperation matches the detection accuracy of always-on inter-fog exchange while reducing its energy by 31-33%, and compressed uploads reduce total energy by 71-95% in matched sensitivity tests. Experiments on three real benchmarks further show that low-overhead hierarchical methods remain competitive in detection quality, while flat federated learning defines the minimum-energy operating point. These results provide practical design guidance for underwater deployments operating under severe acoustic communication constraints.

The autoresearch repository enables an LLM agent to search for optimal hyperparameter configurations on an unconstrained search space by editing the training code directly. Given a fixed compute budget and constraints, we use \emph{autoresearch} as a testbed to compare classical hyperparameter optimization (HPO) algorithms against LLM-based methods on tuning the hyperparameters of a small language model. Within a fixed hyperparameter search space, classical HPO methods such as CMA-ES and TPE consistently outperform LLM-based agents. However, an LLM agent that directly edits training source code in an unconstrained search space narrows the gap to classical methods substantially despite using only a self-hosted open-weight 27B model. Methods that avoid out-of-memory failures outperform those with higher search diversity, suggesting that reliability matters more than exploration breadth. While small and mid-sized LLMs struggle to track optimization state across trials, classical methods lack domain knowledge. To bridge this gap, we introduce Centaur, a hybrid that shares CMA-ES's internal state, including mean vector, step-size, and covariance matrix, with an LLM. Centaur achieves the best result in our experiments, with its 0.8B variant outperforming the 27B variant, suggesting that a cheap LLM suffices when paired with a strong classical optimizer. The 0.8B model is insufficient for unconstrained code editing but sufficient for hybrid optimization, while scaling to 27B provides no advantage for fixed search space methods with the open-weight models tested. Code is available at https://github.com/ferreirafabio/autoresearch-automl.

Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems.

Existing automated research systems operate as stateless, linear pipelines -- generating outputs without maintaining any persistent understanding of the research landscape they navigate. They process papers sequentially, propose ideas without structured gap analysis, and lack mechanisms for agents to verify, challenge, or refine each other's findings. We present \textbf{AI-Supervisor}, a multi-agent orchestration framework where specialized agents provide end-to-end AI research supervision driven by human interests -- from literature review through gap discovery, method development, evaluation, and paper writing -- through autonomous exploration and self-correcting updates of research knowledge. Unlike sequential pipelines, AI-Supervisor maintains a continuously evolving \emph{Research World Model}, implemented as a Knowledge Graph, that captures methods, benchmarks, known limitations, and unexplored gaps, serving as shared memory across all agents and enabling agents to explore and build upon a structured understanding of the research landscape. The framework introduces three architectural contributions: (1) \emph{structured gap discovery} that decomposes methods into core modules, validates their performance across benchmarks, and maps the specific gaps each module creates; (2) \emph{self-correcting discovery loops} that probe why modules succeed on certain problems and fail on others, whether benchmarks carry hidden biases, and whether evaluation protocols remain adequate for emerging challenges; and (3) \emph{self-improving development loops} governed by cross-domain mechanism search that iteratively targets failing modules by finding solutions from other scientific fields. All agents operate under a \emph{consensus mechanism} where independent findings are corroborated before being committed to the Research World Model.

This paper addresses the Service Network Design (SND) problem for a logistics service provider (LSP) operating in a multimodal freight transport network, considering uncertain travel times and limited truck fleet availability. A two-stage optimization approach is proposed, which combines metaheuristics, simulation and machine learning components. This solution framework integrates tactical decisions, such as transport request acceptance and capacity booking for scheduled services, with operational decisions, including dynamic truck allocation, routing, and re-planning in response to disruptions. A simulated annealing (SA) metaheuristic is employed to solve the tactical problem, supported by an adaptive surrogate model trained using a discrete-event simulation model that captures operational complexities and cascading effects of uncertain travel times. The performance of the proposed method is evaluated using benchmark instances. First, the SA is tested on a deterministic version of the problem and compared to state-of-the-art results, demonstrating it can improve the solution quality and significantly reduce the computational time. Then, the proposed SA is applied to the more complex stochastic problem. Compared to a benchmark algorithm that executes a full simulation for each solution evaluation, the learning-based SA generates high quality solutions while significantly reducing computational effort, achieving only a 5% difference in objective function value while cutting computation time by up to 20 times. These results demonstrate the strong performance of the proposed algorithm in solving complex versions of the SND. Moreover, they highlight the effectiveness of integrating diverse modeling and optimization techniques, and the potential of such approaches to efficiently address freight transport planning challenges.

Digital twin technology, when combined with physics-informed machine learning with simulation results of Aspen, offers transformative capabilities for industrial process monitoring, control, and optimization. In this work, the proposed model presents a Physics-Informed Neural Network (PINN) digital twin framework for the dynamic, tray-wise modeling of binary distillation columns operating under transient conditions. The architecture of the proposed model embeds fundamental thermodynamic constraints, including vapor-liquid equilibrium (VLE) described by modified Raoult's law, tray-level mass and energy balances, and the McCabe-Thiele graphical methodology directly into the neural network loss function via physics residual terms. The model is trained and evaluated on a high-fidelity synthetic dataset of 961 timestamped measurements spanning 8 hours of transient operation, generated in Aspen HYSYS for a binary HX/TX distillation system comprising 16 sensor streams. An adaptive loss-weighting scheme balances the data fidelity and physics consistency objectives during training. Compared to five data-driven baselines (LSTM, vanilla MLP, GRU, Transformer, DeepONet), the proposed PINN achieves an RMSE of 0.00143 for HX mole fraction prediction (R^2 = 0.9887), representing a 44.6% reduction over the best data-only baseline, while strictly satisfying thermodynamic constraints. Tray-wise temperature and composition profiles predicted under transient perturbations demonstrate that the digital twin accurately captures column dynamics including feed tray responses, reflux ratio variations, and pressure transients. These results establish the proposed PINN digital twin as a robust foundation for real-time soft sensing, model-predictive control, and anomaly detection in industrial distillation processes.

Current audio deepfake detection has achieved remarkable performance using diverse deep learning architectures such as ResNet, and has seen further improvements with the introduction of large models (LMs) like Wav2Vec. The success of large language models (LLMs) further demonstrates the benefits of scaling model parameters, but also highlights one bottleneck where performance gains are constrained by parameter counts. Simply stacking additional layers, as done in current LLMs, is computationally expensive and requires full retraining. Furthermore, existing low-rank adaptation methods are primarily applied to attention-based architectures, which limits their scope. Inspired by the neuronal plasticity observed in mammalian brains, we propose novel algorithms, dropin and further plasticity, that dynamically adjust the number of neurons in certain layers to flexibly modulate model parameters. We evaluate these algorithms on multiple architectures, including ResNet, Gated Recurrent Neural Networks, and Wav2Vec. Experimental results using the widely recognised ASVSpoof2019 LA, PA, and FakeorReal dataset demonstrate consistent improvements in computational efficiency with the dropin approach and a maximum of around 39% and 66% relative reduction in Equal Error Rate with the dropin and plasticity approach among these dataset, respectively. The code and supplementary material are available at Github link.

Federated Learning (FL) of Large Language Models (LLMs) in multilingual environments presents significant challenges stemming from heterogeneous language distributions across clients and disparities in language resource availability. To address these challenges, we extended the FederatedScope-LLM framework to support multilingual instruction-tuning experiments with LLMs. We also introduced a novel client-specific early stopping mechanism, Local Dynamic Early Stopping (LDES-FL), which allows clients to pause and resume local training based on client-side validation performance, enhancing training efficiency and sustainability. Through a series of experiments, we studied how client language composition - from fully monolingual to increasingly multilingual clients - affects multilingual quality, fairness and training cost. Monolingual local fine-tuning remains the most effective for single-language specialization, whereas federated training is better suited to learning a single balanced multilingual model. In FL, increasing within-client multilinguality leads to stronger and fairer global models, narrows the gap to centralized multilingual fine-tuning, and yields the largest gains for lower-resource languages, albeit at the cost of more optimization steps. Overall, our results identify client language composition as a key design variable in multilingual FL, shaping performance, fairness and efficiency.

Mesh-free numerical methods provide flexible discretisations for complex geometries; however, classical meshless discrete differential operators typically trade low computational cost for limited accuracy or high accuracy for substantial per-stencil computation. We introduce a parametrised framework for learning mesh-free discrete differential operators using a graph neural network trained via polynomial moment constraints derived from truncated Taylor expansions. The model maps local stencils relative positions directly to discrete operator weights. The current work demonstrates that neural networks can learn classical polynomial consistency while retaining robustness to irregular neighbourhood geometry. The learned operators depend only on local geometry, are resolution-agnostic, and can be reused across particle configurations and governing equations. We evaluate the framework using standard numerical analysis diagnostics, showing improved accuracy over Smoothed Particle Hydrodynamics, and a favourable accuracy-cost trade-off relative to a representative high-order consistent mesh-free method in the moderate-accuracy regime. Applicability is demonstrated by solving the weakly compressible Navier-Stokes equations using the learned operators.

Recent advances in large language models (LLMs) have enabled the development of autonomous agents capable of complex reasoning and multi-step problem solving. However, these agents struggle to adapt to specialized environments and do not leverage past interactions, approaching each new task from scratch regardless of their accumulated experience. We introduce Experiential Reflective Learning (ERL), a simple self-improvement framework that enables rapid environment adaptation through experiential learning. ERL reflects on task trajectories and outcomes to generate heuristics, capturing actionable lessons that transfer across tasks. At test time, relevant heuristics are retrieved based on the current task and injected into the agent's context to guide execution. On the Gaia2 benchmark, ERL improves success rate by 7.8% over a ReAct baseline, with large gains in task completion reliability, and outperforms prior experiential learning methods. Through systematic ablations, we find that selective retrieval is essential and that heuristics provide more transferable abstractions than few-shot trajectory prompting. These results demonstrate that reflecting on single-attempt experiences to extract transferable heuristics enables effective agent self-improvement.

The requirement of generating predictions that exactly fulfill the fundamental symmetry of the corresponding physical quantities has profoundly shaped the development of machine-learning models for physical simulations. In many cases, models are built using constrained mathematical forms that ensure that symmetries are enforced exactly. However, unconstrained models that do not obey rotational symmetries are often found to have competitive performance, and to be able to \emph{learn} to a high level of accuracy an approximate equivariant behavior with a simple data augmentation strategy. In this paper, we introduce rigorous metrics to measure the symmetry content of the learned representations in such models, and assess the accuracy by which the outputs fulfill the equivariant condition. We apply these metrics to two unconstrained, transformer-based models operating on decorated point clouds (a graph neural network for atomistic simulations and a PointNet-style architecture for particle physics) to investigate how symmetry information is processed across architectural layers and is learned during training. Based on these insights, we establish a rigorous framework for diagnosing spectral failure modes in ML models. Enabled by this analysis, we demonstrate that one can achieve superior stability and accuracy by strategically injecting the minimum required inductive biases, preserving the high expressivity and scalability of unconstrained architectures while guaranteeing physical fidelity.

Accurate representation of multimodal knowledge is crucial for event forecasting in real-world scenarios. However, existing studies have largely focused on static settings, overlooking the dynamic acquisition and fusion of multimodal knowledge. 1) At the knowledge acquisition level, how to learn time-sensitive information of different modalities, especially the dynamic structural modality. Existing dynamic learning methods are often limited to shallow structures across heterogeneous spaces or simple unispaces, making it difficult to capture deep relation-aware geometric features. 2) At the knowledge fusion level, how to learn evolving multimodal fusion features. Existing knowledge fusion methods based on static coattention struggle to capture the varying historical contributions of different modalities to future events. To this end, we propose DyMRL, a Dynamic Multispace Representation Learning approach to efficiently acquire and fuse multimodal temporal knowledge. 1) For the former issue, DyMRL integrates time-specific structural features from Euclidean, hyperbolic, and complex spaces into a relational message-passing framework to learn deep representations, reflecting human intelligences in associative thinking, high-order abstracting, and logical reasoning. Pretrained models endow DyMRL with time-sensitive visual and linguistic intelligences. 2) For the latter concern, DyMRL incorporates advanced dual fusion-evolution attention mechanisms that assign dynamic learning emphases equally to different modalities at different timestamps in a symmetric manner. To evaluate DyMRL's event forecasting performance through leveraging its learned multimodal temporal knowledge in history, we construct four multimodal temporal knowledge graph benchmarks. Extensive experiments demonstrate that DyMRL outperforms state-of-the-art dynamic unimodal and static multimodal baseline methods.

The Industrial Internet of Things (IIoT) introduces significant security challenges as resource-constrained devices become increasingly integrated into critical industrial processes. Existing security approaches typically address threats at a single network layer, often relying on expensive hardware and remaining confined to simulation environments. In this paper, we present the research framework and contributions of our doctoral thesis, which aims to develop a lightweight, Machine Learning (ML)-based security framework for IIoT environments. We first describe our adoption of the Tm-IIoT trust model and the Hybrid IIoT (H-IIoT) architecture as foundational baselines, then introduce the Trust Convergence Acceleration (TCA) approach, our primary contribution that integrates ML to predict and mitigate the impact of degraded network conditions on trust convergence, achieving up to a 28.6% reduction in convergence time while maintaining robustness against adversarial behaviors. We then propose a real-world deployment architecture based on affordable, open-source hardware, designed to implement and extend the security framework. Finally, we outline our ongoing research toward multi-layer attack detection, including physical-layer threat identification and considerations for robustness against adversarial ML attacks.