Papers

Frontier LLMs can perform multi-step reasoning over content-free filler tokens like dots or counting sequences, producing correct answers with no visible chain-of-thought (CoT). This is a limit case for behavioral oversight, where surface tokens carry no information about the underlying reasoning. But hidden from the output is not the same as hidden from us. On four task families (fact retrieval, parallel numeric composition, string manipulation, and in-context computation), two open-weights frontier models (DeepSeek V3, Kimi K2) compute over filler tokens in a legible way: attention routes the question through the filler region to the answer, logit-lens readouts show retrieved facts emerging early and their composition crystallizing in late layers, and KV-cache transplants at filler positions causally swap outputs between examples. We introduce an unsupervised decoding pipeline that takes only hidden states as input and recovers intermediate values with 82-94% accuracy (best LLM judge) across both models and all four tasks, without ground-truth labels or training. Even without a judge, the hidden values are already directly in the pipeline's top-2 tokens 35-85% of the time. The uplift persists whether the filler is prefilled or the model generates the filler itself. On these cleanly decomposable tasks, hidden computation that defeats behavioral CoT monitoring is readable from the residual stream, which suggests that monitorability is a property of the model's full computational trace rather than only its surface tokens.

Copernicus, the European Union's Earth observation program, produces petabytes of Earth observation and climate data, offering immense potential for research, policy, and applications. However, access to these datasets requires advanced programming skills and familiarity with domain-specific formats such as NetCDF or GRIB. Moreover, general-purpose Text-to-SQL systems fail when applied naively to the meteorological domain due to a profound “Symbolic-to-Numeric” gap. To overcome these limitations, we present an end-to-end Text-to-SQL framework specifically engineered for real-world, scalable meteorological data exploration. Our system intercepts natural language to resolve spatial and semantic ambiguities before SQL generation. We design STRATOS, a Spatio-Temporal Resolution Agent for Text-to-SQL to dynamically bridge the symbolic-to-numeric gap by mapping fuzzy concepts to a localized ontology and resolving spatial entities via external knowledge bases. Further, our complexity-aware query rewriter rewrites expensive spatial predicates, reducing execution times from hours to seconds. Last, we introduce the STRATOS Evaluation Workload, comprising 7,520 complex query pairs explicitly designed by domain experts to test scalability and symbolic-to-numeric translation across challenging spatio-temporal dimensions previously unexplored by Text-to-SQL systems.

Mutual information estimation is a central problem in machine learning and statistics; however, existing benchmarks typically evaluate estimators on simplified, low-dimensional distributions, leaving their performance on complex, realistic data largely unexplored. We address this gap with a comprehensive benchmarking framework grounded in a unified copula-theoretic perspective that subsumes existing benchmarks as special cases. Within this framework, we propose two complementary families of tests: a copula-first family that systematically varies ground-truth mutual information, dimensionality, and marginal complexity using synthetic and flow-based transformations; and a marginals-first family that couples real-world image data with controlled dependency structures, extending the classic same-class-pairing paradigm. We use this suite to extensively evaluate three classes of estimators: non-parametric, variational, and generative. Contrary to prevailing assumptions, our results indicate that there is no universal winner: each category can systematically outperform all other estimators under specific setups. By analyzing these cases, we identify fundamental estimation barriers and propose new tests that more effectively stress these specific limitations. We share the open source code at https://github.com/VanessB/mutinfo.

Post-training of frontier language models is conducted on curated task suites, and inevitably leaves a distribution shift between training and deployment environments. This exposes developers to generalization failures, which are relatively poorly understood. To better understand such generalization failures, we believe the community should construct clean demonstrations under simplified conditions. To facilitate this, we propose a simple and flexible way to construct language models which fail to generalize in controllable ways when subsequently trained with Reinforcement Learning (RL) on a given distribution of training tasks. Our construction uses Supervised Fine-Tuning on a dataset of a mixture of transcripts corresponding to a collection of 'conditional policies', which can each independently be assigned certain behaviors on each different task distribution, to obtain a model that is then well approximated as a 'mixture of conditional policies.' We observe that RL training then selects for policies that obtain the highest reward on the training distribution. This can produce striking behaviors: in a controlled setting with two distributions containing identical questions prepended with two different 'trigger strings', RL training on either distribution actively degrades performance on the other to zero, even though the underlying task is identical. We also use our construction to illustrate two novel ways in which generalization may fail in future language models, corresponding to distribution shifts of task coverage and temporal context respectively. While our construction is deliberately simple and may not closely resemble 'natural' generalization failures, the resulting 'model organisms' are of interest for alignment stress-testing and generalization science, and can be used as existence proofs that training success and generalization can come apart in structured ways.

This study aims to predict Tumor, Node, and Metastasis (TNM) stage labels independently, with the Cancer Genome Atlas (TCGA) pathology report as the sixth shared task of SMM4H-HeaRD 2026. The problem is framed as three multi-label classification tasks. We explore both classical and deep learning approaches using Term Frequency-Inverse Document Frequency (TF-IDF) features and embeddings from ClinicalBERT, BioBERT, and PubMedBERT. These representations are used with Logistic Regression (LR), Light Gradient Boosting Machine (LightGBM), Feed-Forward Neural Networks (FFNN), and Wide Residual Networks (WRN). Our results show that individual embeddings perform similarly to the TNM label classification, while their combination improves its predictive ability. WRN achieves AUROC scores of 0.839 (T), 0.8502 (N), and 0.803 (M) with F1-scores of 0.622, 0.702, and 0.9337, respectively, for the training phase. LightGBM with TF-IDF performs best with AUROC scores of 0.9368 (T), 0.9524 (N), and 0.8311 (M) and F1-scores of 0.7559 (T), 0.7384 (N), and 0.7017 (M) during the training phase. Furthermore, the result of the Codabench for the test sets indicates a Macro-F1 score of 0.978, 0.957, and 0.879 for the T, N, and M categories respectively for test set 1; while test set 2 records a Macro-F1 score for T, N, and M is 0.807, 0.767, 1.0 respectively. However, performance declined during the evaluation phase of the test sets, a drop from 0.938 to 0.858 of test set 1 to 2, for the Macro-F1 score across all stages; suggesting limitations in model generalizability, sensitivity to class imbalance, and challenges in processing lengthy clinical documents. Although this study provides an efficient baseline model and a reproducible pipeline, further optimization and validation are required before it can be considered suitable for use in a real-world clinical setting.

Inference-time alignment methods, such as Best-of-$N$, offer a flexible alternative to training-based alignment by using reward models to select high-quality responses generated by a reference LLM. However, the efficacy of these methods is inherently limited by the response quality: if the reference LLM assigns negligible probability to high-reward responses, no selection strategy will succeed in finding aligned outputs. In this work, we propose Best-of-Better-$N$ (BoBN), an in context learning-based generation framework to address this challenge. Our method utilizes retrieval from high-reward examples relevant to the input query and task. Crucially, we introduce a restyling step where retrieved responses are rewritten by the reference LLM to align with the target task's format and style. These restyled examples are used in-context to shift the sampling distribution toward the high-reward region. We analytically characterize how in-context learning shifts the output distribution of pretrained transformers toward the high-reward region, resulting in provable benefits on the target task. We then evaluate BoBN on safety alignment and mathematical reasoning benchmarks across several reference LLMs. BoBN's higher-quality responses enable better performance to be achieved when the number of responses $N$ is fixed, and smaller $N$ required to achieve a target performance.

While skill optimization for autonomous agents has gained traction, existing methods rely on complex pipelines. This leaves a fundamental question unaddressed: What constitutes a minimal viable pipeline for skill optimization, where every component is justified by theory or empirical necessity? We formalize skill optimization via Zeroth-Order (ZO) optimization, mapping classical counterparts (central difference, trust regions) to recent literature. Noting that unlike blind numerical perturbations in classical ZO, skill trajectories serve as interpretable debugging feedback. Grounded in Claude Code philosophy and PAC learning, we establish three principles for convergence and generalization: file-system-based trajectory exploration, consensus attribute mining, and independent validation gating. Eliminating redundancies, we propose SkillOpt-Lite. It accelerates convergence and outperforms full SkillOpt: improving LiveMath by +8.8 points on GPT-5.5 and +25.4 points on GPT-5.4-nano, allowing the nano model to surpass standard GPT-5.4 optimized by SkillOpt. Finally, we integrate our framework into production coding agents like VSCode Copilot, enabling developers to evolve agent skills via one line of vibe. Because our framework treats all agent components simply as standard editable code, this minimal pipeline naturally generalizes to full harness optimization (HarnessOpt). On SpreadsheetBench, HarnessOpt enables GPT-5.4-nano to achieve 0.7758 accuracy, outperforming the larger GPT-5.5 running standard pipelines (0.7620). Code is available at https://github.com/EvolvingLMMs-Lab/SkillOpt-Lite.

Current Vision-Language-Action (VLA) models excel at robotic manipulation but often struggle with non-Markovian tasks requiring long-term memory and reasoning due to their reliance on immediate observations. Existing solutions face a ”frequency-competence paradox,” where stronger reasoning models are too slow for real-time control, while faster models lack sufficient reasoning capabilities. To resolve this architectural misalignment, we propose HiMe, a Hierarchical Embodied Memory framework that decouples embodied intelligence into a high-frequency Executor for execution, a Sentry for working memory, and a Planner for long-term strategy. We also introduce a dynamic knowledge system based on cross-modal semantic schemas and active management mechanisms, allowing robots to maintain memory plasticity through ”Add, Update, and Delete” operations. This hierarchical design effectively balances the conflict between real-time execution and slow thinking planning, significantly improving success rates in long-horizon tasks. Experiments demonstrate that this approach not only outperforms flat memory baselines but also exhibits the novel ability to self-correct its internal knowledge based on human preferences.

Knowledge graphs (KGs) that underpin Graph-based Retrieval-Augmented Generation (Graph-RAG) are increasingly built automatically by LLM-driven extraction rather than curated by experts. Proper evaluation would require instrumenting all pertinent stages: extraction, graph construction, and inference, coherently enough to localize failures, so that a failure at one stage is not discovered as a wrong answer at the end. We introduce TRIAGE, a stage-aware instrumentation framework for automated, document-grounded graph-RAG that asks not only whether the underlying graph can be trusted but at what cost it can be queried. TRIAGE attaches stage-specific, independently interpretable metrics to three stages: the KG Implementation (triple confidence, source coverage, and schema and canonicalization checks), the KG Validation by expert (graph-level structural quality, with correctness and completeness computed only as offline calibration when a reference is available), and the KG Usage (retrieval coverage, faithfulness, and retrieval cost); the deployed metrics need no gold annotations, the gold-requiring ones serving only as offline calibration. At usage time these metrics form a diagnostic chain of necessary conditions whose first broken link localizes the failure, and the diagnosis maps to the stage levers that can remedy it: extraction, graph and schema, or retrieval. TRIAGE is a theoretical framework with a proof of concept and a reproducible evaluation protocol.

LLM agents often degrade over long episodes: as trajectories grow, they revisit explored states, repeat failed actions, and lose strategies that previously worked. Test-time training (TTT) offers a way to adapt model weights to the evolving task state, but existing LLM TTT methods largely adapt once to a fixed input. We study continuous TTT in multi-turn agent episodes, where each update changes the policy that generates later training text. This creates a self-training loop that helps when new trajectory information appears, but can amplify drift when the agent gets stuck and repeatedly trains on similar text. We find that update-text repetition distinguishes these regimes and introduce Agentic Test-Time Training (aTTT), a token-level reweighting method that downweights the loss on tokens appearing in repeated $n$-grams from prior updates while leaving novel tokens fully weighted. To run such updates inside live episodes, we build a concurrent serving system using vLLM's runtime LoRA API, limiting overhead to 1.9$\times$ the no-TTT cost. aTTT improves success by up to 5.0 points on ALFWorld and 4.9 points on SWE-bench Lite. The gains concentrate where models already have task competence but drift over long trajectories, suggesting that aTTT mainly preserves existing competence rather than teaching new abilities.

Large language models (LLMs) exhibit strong reasoning and world-knowledge capabilities, yet often struggle to gather information effectively across the multi-turn interactions required in sequential decision-making settings. We introduce Amortised Sequential Information Gathering (ASIG), a fine-tuning approach that amortises Bayesian Experimental Design (BED) into LLM policies via a multi-turn extension of Group Relative Policy Optimisation with an Expected Information Gain reward. Evaluated on the 20 Questions task, ASIG more than doubles the success rate of the 7B base model and reduces inference cost by over $25\times$ relative to BED-LLM, a competitive inference-time baseline. Applied to MediQ, a medical diagnosis benchmark unseen during training, ASIG improves information-seeking performance at the 7B scale, suggesting that the learned strategies can transfer out of distribution. Our findings show that amortising BED into LLM policies provides an effective and computationally efficient approach to sequential information gathering.

Algorithmic recourse addresses the challenge of providing tailored recommendations to users affected by unfavorable machine learning decisions, in potentially high-stakes scenarios. Traditional approaches to recourse often rely on the closest counterfactual explanations or assume a priori knowledge of a user's causal structure, resulting in interventions that overlook individual contexts and specific feature interactions. To overcome these limitations, we study a human-in-the-loop framework that iteratively approximates the user's structural causal model through interactive queries via Bayesian inference before producing recourse recommendations. This framework exploits humans' feedback to improve the identification of causal effects, allowing personalized recourse that is plausible, cost-effective, and aligned with the actual causal dependencies of each user. As a proof of concept, we evaluate this framework through simulated human responses. Our simulations across linear and non-linear causal models show promising results, though challenges remain in capturing complex, non-linear structures, emphasizing the importance of accurate approximations and robust noise distribution modeling.

Modern AI agent implementations such as frontier coding agents chain multiple tools at runtime that create a security surface that per-tool guardrails are unable to address, as individually permitted tools can violate organizational policies when composed. We propose the Dynamic Security Control Compositor (DSCC), a two-phase approach to compositional security for multi-tool agent chains. In Phase 1, at session checkout, a Most Restrictive Set (MRS) algorithm composes per-tool security policies into a single effective policy with a formal monotonicity invariant that extending a chain can only tighten the result, blocking incompatible combinations before any tool executes. Outputs of any tool call propagate their classification constraints into a session-level taint state, so subsequent invocations must satisfy the most restrictive constraints seen so far. In Phase 2, at runtime, the system tracks the sensitivity of data the agent touches through a monotonic taint state and revokes the session if the accumulated exposure would make a subsequent tool call a policy violation. Together, these phases provide defense in depth, where static composition prevents unsafe chains from starting, and runtime taint tracking catches violations that emerge from the specific data used. We provide a reference implementation on 32 tools governed by 16 NIST SP 800-53 aligned policies and evaluate it under two composition modes. In the default clearance mode, permitted combinations are partitioned into classification-level clusters, blocking 79.2% of policy pairs and 95.5% of triples. The alternative taint mode admits mixed-classification chains within the exfiltration boundary, blocking 42.5% and 60.5% respectively. We discuss the governance implications for organizations deploying multi-tool agents, including the utility-security tradeoff and the changes needed to operationalize chain-aware policies.

A trustworthy and GDPR-compliant deepfake audio detector must base its decisions on acoustic artifacts, not on what is being said or who is speaking. We present a large-scale study of semantic independence for Resemble AI's detector, DETECT-3B-Omni. Using 10,240 audio samples from diverse US English speakers across 30 states, generated through 8 different AI voice-cloning systems, we test whether detection accuracy depends on spoken content (benign versus malicious), speaker gender, speaker age, or speaker region. Using equivalence testing, our results show that the accuracy difference between any two of these groups is at most 2 percentage points, at 99% confidence. The detector therefore identifies AI-generated audio with equivalent accuracy regardless of what the audio says or who the speaker is.

Jamming and spoofing threaten wireless and satellite navigation by disrupting or manipulating radio frequency (RF) signals, undermining availability, integrity, and trust. Robust interference monitoring (i.e., detection, classification, characterization, and direction finding) is therefore essential to identify and localize anomalous signals. While machine learning (ML) promises improved performance in complex environments, its development and validation depend on large-scale datasets that capture realistic signal and channel variability. Collecting such data in the real world is difficult because intentional jamming is illegal and ground-truth attribution is confounded by propagation, hardware, and environmental effects. To address this gap, we create and publish S-ICDF, a large-scale indoor interference dataset generated with Sionna, a GPU-accelerated simulation library for physical-layer wireless communications. S-ICDF covers 102 interference configurations, including diverse antenna array patterns, bandwidths, and simulation settings such as noise level and reflection depth. We further provide baseline results by benchmarking S-ICDF with classical estimation and direction finding (DF) methods (MUSIC, ESPRIT, and CAPON) and with modern ML approaches. The dataset is publicly available at: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/sicdf_dataset

Text-to-Image diffusion models often propagate harmful bias inherited from the training data. Existing bias mitigation techniques typically intervene only at the text encoder or provide inference-time guidance, often leading to generations that collapse into semantically incoherent outputs. To address these limitations, we introduce CO-ALIGN (Concept Ontology Alignment), a novel bias mitigation approach based on concept-graph alignment that operates on the model's internal concept ontology. By aligning concepts within the text encoder and denoiser, CO-ALIGN achieves substantial bias reduction while preserving generative integrity. We demonstrate the effectiveness of concept-graph alignment across three paradigms: text-encoders, denoisers and joint text-denoiser ontology alignment. CO-ALIGN outperforms the state of the art, improving fairness by $30\%$, $\Delta FID=11.4$ in image quality, $2.8\%$ in image fidelity, all while reducing semantically incoherent outputs by $88\%$. Beyond bias mitigation, we show that CO-ALIGN benefits other downstream tasks as well. In particular, our experiments demonstrate that better-aligned internal ontologies enhance concept unlearning robustness across multiple unlearning techniques.

Real time location data derived from mobile applications is a powerful tool for addressing various urban challenges, including tourism planning, parking management, bus route optimization, and resource allocation. Besides, it offers invaluable insights for shaping strategic decisions in commercial domains such as location based services, market share analysis, and behavioral profiling. In this expansive study, we aim to address all of the aforementioned challenges by investigating the behaviors and patterns of smartphone users within urban environments, particularly in the domains of tourism, transportation, and retail. Our approach encompasses the development of a sophisticated data platform from inception to implementation, which includes the formulation of use cases, architectural design, and implementation of modules. We employ state of the art techniques and technologies, including data anonymization, ETL pipelines, and utilizing Google BigQuery and Vertex AI for data processing and machine learning model development. A modular architecture based on reusable analytical building blocks was developed to generate data products that support multiple stakeholder driven use cases. Additionally, we apply interactive data visualization techniques via Power BI to facilitate the effective interpretation of analytical findings by stakeholders. The developed models address a wide range of mobility analytics tasks, including mobility profiling, frequent trajectory mining, area of influence analysis, traffic anomaly detection, and origin destination pattern analysis. The results demonstrate the framework's ability to capture user mobility dynamics at fine spatial and temporal resolutions, providing actionable insights for urban planning and strategic business decision making.

Agentic AI systems are increasingly used to edit, refine, and repair decision policies, but evaluating these edits is difficult when per-state expert action labels are unavailable. We study this problem in a hotel-pricing simulator where an agentic policy editor receives only region-level diagnostic feedback: summaries of how its price distribution differs from a benchmark policy across time, inventory, and market regions. The editor cannot observe benchmark actions, benchmark source code, reward numbers, or held-out outcomes, and may only propose constrained edits to a target-action table. On 5,000 held-out episodes, a multi-restart LLM editor reaches RevPAR 108.47 (95% CI 107.61 - 109.34), close to the benchmark policy's 108.75 (107.81 - 109.68), with paired gap (LLM minus benchmark) -0.276 and 95% CI [-0.692, 0.146]. A cheap diagnostic projection already recovers much of the revenue (107.90), so the LLM editor's distinctive gain is not raw revenue lift alone: it also reduces episode composition distance from 1.153 to 0.609. This is the strongest non-benchmark repair result. This profile is not explained by restart search alone: non-semantic proposers with up to 2,500 evaluations fall 8.77 - 14.57 RevPAR points short. Nor is it explained by plausible prompt format: a shuffled-diagnostic control breaks region-error correspondence and falls to RevPAR 94.30. The match is genuine but partial. A tree editor achieves stronger pooled alignment, 0.214 versus 0.266, and stronger reference-state D1, 0.328 versus 1.197, yet revenue falls to 98.91. These results show that agentic policy repair should be evaluated by whether diagnostic feedback becomes reliable closed-loop outcome, not by a single behavioral distance.

The rapidly growing repository of publicly available large language models (LLMs) presents significant challenges for systematic management and quantification at scale, such as model lineage tracing, licensing, and evaluation. However, task-specific benchmarks are insufficient for this setting, as LLMs differ widely in architectures, scales, and training procedures. To address this challenge, we adopt spectral shape-based metrics for managing and quantifying LLMs based on Heavy-Tailed Self-Regularization theory. Our approach uses the shape information of the weight empirical spectral density as a compact spectral signature of each model. This signature captures intrinsic properties of pretrained models and remains robust during post-training, making it suitable for model-level analysis. In addition, this metric is data-free, computationally-efficient, and scale-invariant, enabling large-scale analysis in practice. Moreover, we curate a large and diverse model corpus consisting of major open-source LLM families, and use it to systematically benchmark spectral and non-spectral metrics across models and downstream tasks. We show that our spectral signature supports the tracking of the model lineage, the unsupervised clustering of similar models, and the quantification of the model performance. Overall, the proposed spectral signature provides a meaningful proxy for broad performance trends across LLMs, enabling efficient organization, comparison, and analysis of large model collections.

Vision-language models (VLMs) increasingly read news and web content as images, where the publisher's identity is visually present. We show that VLMs carry a strong source-credibility prior keyed on outlet identity, and study it along three axes. (i) Cross-model benchmark. We introduce CueTrust, a cross-model diagnostic that measures which surface source cue overrides an article's content evidence via a Source-Override Index (SOI). Across seven VLMs and five cues, the vulnerability profile is model- and scale-dependent, and the override is outlet-identity-specific and encoding-invariant, firing from the masthead name, the logo image, or the bare domain, but not from a named author, in-text authority, or page layout (clean negative controls). (ii) Mechanistic account. For the brand cue, we give a full mechanistic account: swapping only the masthead moves credibility across an approximately 11 log-odds range that tracks professional ratings (rho = 0.88 with Media Bias/Fact Check). The prior is dual-coded (name and logo), strengthens with scale, is causally formed at layers 19-21, carried by interpretable seed-stable sparse-autoencoder features, and recurs at the same relative locus in a second model family. It overrides content (about 1.8x) as a signal-magnitude effect within a shared pathway, not a privileged route. Steering the localized direction selectively reduces the override (41% reduction) and generalizes to held-out outlets, confirming the prior is causally used, not merely decodable. Deployed VLMs may thus defer to source identity over the evidence in front of them, a reliability failure we can measure across models, localize, and causally probe. We release the stimulus suite and CueTrust.

Malicious Python packages have become a major threat to software supply chain ecosystems due to the widespread adoption of open-source repositories such as PyPI. Existing learning-based detection methods struggle to capture the hierarchical organization and heterogeneous interactions among different program entities. Although Large Language Models (LLMs) have demonstrated strong capabilities in code understanding and semantic reasoning, they are rarely integrated with structural program representations for fine-grained malicious behavior analysis. In this paper, we propose an LLM-enhanced hierarchical heterogeneous graph representation learning framework for malicious Python package detection. The framework constructs a hierarchical heterogeneous code graph that explicitly models heterogeneous code entities and different types of structural dependencies. LLMs are further leveraged to infer function-level semantic roles, introducing an additional layer of semantic heterogeneity. Based on this graph, we develop a hierarchical heterogeneous graph neural network that performs type-aware message passing over different node and edge categories, effectively modeling malicious behavior propagation for accurate package-level classification. The framework also incorporates a function-level attribution mechanism which, combined with LLM reasoning, automatically identifies suspicious functions and localizes fine-grained malicious behaviors without human expert intervention. Extensive experiments on real-world datasets show that our framework consistently outperforms traditional machine learning methods, graph-based detectors, and state-of-the-art LLMs across packages with varying sizes and dependency complexities, while providing accurate, robust, and interpretable malicious behavior localization.

The accuracy of consumer-grade inertial navigation is bottlenecked by the stochastic noise of Micro-Electro-Mechanical Systems (MEMS). Traditional deterministic neural architectures often succumb to “estimation jittering,” sacrificing high-frequency kinematic fidelity for numerical stability. We propose PedestrianDiffusion, a multimodal spectral-domain generative framework reformulating dense 6D state estimation as a continuous conditional denoising process. By operating in the frequency domain, our formulation bounds the spectral covariance, acting as a mathematical preconditioner to stabilize the reverse diffusion trajectory. Furthermore, we introduce a zero-shot semantic conditioning mechanism leveraging vision-language embeddings as categorical priors to generalize across heterogeneous sensor noise profiles. To address the computational intractability of generative tracking, we deploy a single-step deterministic probability flow ODE solver ($T=1$). This yields high-capacity asynchronous batch trajectory refinement, establishing the viability of generative architectures for asynchronous batch trajectory refinement on edge hardware. Extensive evaluations on the OxIOD, RIDI, RoNIN, and TLIO benchmarks demonstrate that PedestrianDiffusion achieves state-of-the-art performance, exhibiting unprecedented robustness to impulse perturbations and coupled 6D kinematic drift. This work provides a rigorous algorithmic blueprint for next-generation Neural Inertial Measurement Units (N-IMUs).

Accurate and efficient dataset valuation is essential for enabling fair and transparent data marketplaces, especially when multiple contributors provide data for training multi-task models. Most existing valuation methods, however, are limited to single-task settings, overlooking scenarios where a buyer aims to optimize performance across multiple downstream tasks. Moreover, traditional valuation approaches, such as Shapley-based or retraining-based methods, are computationally expensive and poorly suited for decentralized environments without a trusted central coordinator and with strict privacy constraints. We propose DMVM (Decentralized Multi-task Valuation via Model Merging), a novel framework that bypasses retraining and data sharing by leveraging task arithmetic to infer dataset contributions directly from model combinations. Instead of retraining or sharing raw data, DMVM quantifies how models trained on different datasets combine in parameter space to infer each dataset's marginal utility across multiple tasks. This formulation yields a valuation process that is scalable, computationally efficient, and explicitly aligned with multi-task generalization behavior. To support decentralized deployment, we introduce a secure aggregation protocol that enables collaborative valuation without revealing individual model parameters or private data. We also provide theoretical error bounds characterizing the approximation quality of DMVM and validate our framework through comprehensive experiments on computer vision and natural language processing tasks.

The federated learning (FL) paradigm fosters distributed pervasive computing combined with artificial intelligence techniques, allowing for optimized data usage and improved mitigation of privacy concerns. Indeed, model training occurs on the client's local devices, and model parameters are subsequently shared with a centralized server. However, there is a need to find a tradeoff between models' personalization and generalization capabilities. In this paper, we design and implement several testing scenarios devoted to evaluating and comparing the centralized, local, and federated paradigm performances. We also design and implement a scenario that emulates a change in clients' data. We then present experimental results of the FedAvg algorithm applied to the Human Activity Recognition (HAR) domain to understand the trade-off between personalized and generalized accuracy. Results show that, although FedAvg confirms a higher degree of personalization capabilities while keeping a high degree of generalization with respect to the traditional centralized learning, this result is not so obvious under stressful conditions, such as when varying class distribution over clients.

LLM agents are becoming a common interface for research, coding, and question answering, yet their Thought-Action-Observation loop is often serial: the model reasons, emits a tool call, then idles the GPU until the result returns. This wait consumes 16-37% of wall time in our workloads and 35-61% in prior reports. Speculative tool execution can hide this wait, but existing systems need auxiliary predictors, historical traces, or static workflow graphs, leaving a gap for training-free, day-one deployment. We observe that the model can be its own predictor: a probe forked at the start of generation predicts Qwen3-32B's upcoming tool name with 74.6-99.6% accuracy across five benchmarks. We present SPORK (Self-sPeculative fORKing), a training-free controller that dispatches the speculated tool call early, overlapping its execution with the remaining chain-of-thought decode. A cost model captures when speculation breaks even, and each component improves one of its terms: a prefix-cache fork cuts probe cost, a confidence gate filters mispredictions, and partial-token accept turns rejected probes into speculative-decoding drafts. On acceptance, the tool result is ready when reasoning ends; on rejection, SPORK falls back to serial execution with no correctness penalty. On real-tool benchmarks, SPORK cuts Qwen3-32B's GAIA P95 by 18% (131.9 to 108.1 s); the mechanism holds across model sizes from 4B to 32B and across dense and mixture-of-experts models, with task accuracy within 1 pp of baseline or better wherever measured. SPORK deploys as a thin controller over standard completion APIs (no retraining, no auxiliary models, no offline traces) and is orthogonal to token-level speculative decoding. SPORK is open source at https://github.com/baihuajun24/spork.

We present an automated pipeline that decomposes Italian tax-court judgments into individual legal issues and extracts, for each issue, a structured XML representation grounded in the IRAC framework and the legal syllogism. The pipeline targets a corpus of approximately $330{,}000$ first- and second-instance decisions of the Italian tax courts and is built around a capable yet cost-efficient general-purpose model (DeepSeek V3), a choice driven by the need to process several hundred thousand documents at a sustainable cost. To address the well-documented unreliability of large language models on legal citations, we couple the extraction step with an automatic hallucination-detection filter that compares the references produced by the model with those identified in the judgment text by a dedicated parser (Linkoln), normalised to standard identifiers (URN-NIR, ECLI, CELEX). We validate the pipeline on $50$ judgments annotated by two PhDs in tax law, computing inter-annotator agreement and LLM-vs-expert agreement on both issue extraction and legal citations, together with a stand-alone evaluation of the hallucination filter. To the best of our knowledge, this is the first issue-level, expert-validated structured extraction pipeline with hallucination control for Italian tax-court decisions, and it provides a concrete starting point for downstream applications such as issue-level retrieval, citation-network analysis, and the construction of large-scale datasets of legal reasoning.

Eco-friendly energy management for artificial intelligence data centers (AIDCs) is crucial because of the significant increase in energy consumption-induced carbon emissions from AIDCs resulting from the rapid expansion of AI applications. This paper proposes a hierarchical carbon-aware multi-agent reinforcement learning (CA-MARL) framework for robust and efficient operations of AIDCs under uncertainties while ensuring low-carbon operation of power distribution systems. The framework comprises a workload manager (WM) agent and multiple local AIDC agents trained using a multi-agent transformer method, corresponding to a global AIDC aggregator and a local AIDC operator, respectively. Leveraging AIDC operation data along with nodal carbon intensity (NCI) calculated from the carbon emission flow-integrated distribution system operator problem, the WM agent spatially allocates AI training and inference jobs among all AIDCs. Based on the jobs allocated from the WM agent and NCI information, each AIDC agent schedules economical and eco-friendly operations of the AIDC by performing the following tasks: i) temporal shifting of training jobs, ii) spatial allocation of training graphics processing unit (GPU) blocks and inference GPUs within the AIDC, and iii) control of the supply air temperature of the cooling system. The effectiveness of the proposed framework was assessed using an IEEE 33-node power distribution system.

Agentic code review, where autonomous agents provide code review comments on pull requests, is increasingly integrated into development workflows, yet there is limited empirical evidence on how developers respond to such comments in practice. In this paper, we present an empirical study of agentic code reviews using CodeRabbit as a case study. Through an empirical study of 31,073 pairs of code reviews and developer feedback from 10,191 pull requests across 239 GitHub repositories, our results show that agentic reviews receive mixed reception: 36.4% were accepted and 7.3% triggered discussion, while 56.3% were rejected. Rejections were primarily associated with invalid suggestions that were false positives, redundant, or out of scope, as well as misalignment with developer intent and coding practices. We further found that agentic reviews tend to focus more on functional concerns than evolvability-related comments, yet they were more likely to be invalid. To improve effectiveness in review practices, we explored various LLM-based approaches for predicting review rejection. We found that lightweight learning-based methods achieve up to 76% F1 score, suggesting learnable patterns exist between code reviews and their corresponding feedback. Our results highlight the current state of CodeRabbit's agentic code reviews, showing opportunity gaps for improvement, as well as shortcomings hindering its effectiveness.

While Large Language Models (LLMs) can provide personalized support in learning, several studies have raised concerns regarding their use in education. Importantly, learning depends on how students engage with LLMs. This study examined how two types of LLM-based tutors shape students' prompting practices, learning, and subsequent LLM-use: a Socratic-Guidance (SG) tutor, which structures interaction through dialogic questioning, and a Prompt-Refinement (PR) tutor that guides the formulation of effective prompts. We conducted a two-phase study in a graduate-level mobile robotics course: 66 students used either the SG or PR tutor during a 6-week intervention, followed by 52 students using an unconstrained LLM during a 3-week course project. Results show that while the SG- and PR tutors led to similar task performance and prompting patterns during guided use, they differ in learning outcomes and later LLM-use. SG-students, relative to PR-student, achieved higher learning gains in later sessions, and were more likely to adopt understanding-driven prompting strategies, which are predictive of higher understanding, when using an unconstrained LLM. Although learners perceived the SG tutor as less efficient, the findings suggest that Socratic guidance supports the development of students' capacity to learn with LLMs over time, highlighting its importance for LLM tutor design.

Foundation models for Earth systems have so far been trained primarily on physical climate and weather data, with limited representation of the human systems that both drive and respond to environmental change. The lack of a unified global training resource that combines climate, land, ocean, cryosphere, infrastructure, hazards, and socioeconomic data on a common grid hinders progress toward truly multimodal Earth system foundation models. We present WorldTensor, a harmonised global dataset that aligns hundreds of environmental and socioeconomic variables to a standardised 0.25$^\circ$ spatial grid and annual temporal framework. WorldTensor integrates reanalysis products, remote sensing, emissions inventories, land use reconstructions, hydrological observations, infrastructure and hazard datasets, and socioeconomic indicators within a single representation designed for machine learning workflows. To build the dataset, we regridded inputs across heterogeneous native resolutions and projections, rasterised point and vector datasets into spatially meaningful gridded fields, and reconciled temporal coverages ranging from daily observations to sparse multiyear socioeconomic snapshots. All outputs are distributed as NetCDF files with standardised coordinates, variable metadata, and a common CF metadata convention. WorldTensor provides a reproducible resource for training and evaluating foundation models that learn coupled dynamics across environmental and human systems at planetary scale.