Agregador

arXiv:2610.29134v1 Announce Type: new Abstract: Full fine-tuning has achieved remarkable success on reasoning-intensive tasks, but it often requires substantial GPU resource consumption. Parameter-efficient fine-tuning (PEFT) provides a promising alternative for reducing this cost, yet we observe a substantial performance gap on reasoning-intensive tasks. In this paper, we present three key observations showing that this gap is closely associated with norm-related under-adaptation, where PEFT produces markedly smaller weight-update norms than full fine-tuning. Based on these observations, we propose Norm-Guided Parameter-Efficient Fine-Tuning (NormTune), a simple yet effective strategy that adaptively strengthens insufficiently updated layers according to their weight-update norms. Extensive experiments show that NormTune consistently enhances multiple PEFT methods across challenging reasoning benchmarks. Specifically, when trained on S1K-1.1, integrating NormTune into rank-8 LoRA, Bone, and MoRA yields substantial average accuracy improvements of 24.24\%, 11.33\%, and 16.21\% , respectively, across AIME 2024, MATH 500, GPQA, and AIME 2025. Moreover, NormTune-enhanced rank-256 LoRA, Bone, and MoRA achieve performance comparable to S1 full fine-tuning, with respective average gains of 2.17\%, 0.93\%, and 3.00\% over this baseline.

arXiv:2610.29122v2 Announce Type: new Abstract: Individual patient data are the gold standard for secondary analysis of time-to-event evidence, but they are rarely released, and reconstruction from a published Kaplan-Meier figure could help. Existing methods all claim high fidelity, each judged by whichever agreements it happens to achieve, and each returns a single dataset although the map from figure to data is not bijective. We define fidelity explicitly: a dataset is admissible when it reproduces the risk table exactly, follows the digitized curve to within the image resolution, and respects the printed median, interval and event count. We propose StochIKM, a stochastic algorithm that digitizes the figure with a vision-capable language model and uses the high-fidelity definition as constraints, returning an admissible set of individual patient datasets. A solution is guaranteed to exist whenever the figure has been read correctly, so a failure indicts either the published figure or its digitization. In simulation and in 79 arms from published oncology trials, StochIKM reproduces the published risk tables and medians more closely than LLM-improved IPDfromKM, a high-impact alternative.

arXiv:2610.29121v1 Announce Type: new Abstract: A coding agent that rewrites its own scaffold has to choose which proposed edit to expand before that edit has been evaluated, and the search then has to return one agent. Each run fits one linear readout on the parent's steering-split solve rate and eight coordinates of the proposed diff, from 96 scaffold edits labeled by four evaluations of a 16-instance fit split, and then freezes the coefficients. HumanEval+ reserves 48 instances as the steering split and Polyglot reserves 64. That split scores parents and keeps an archive of four nodes. Huxley-G\"odel Machine selection draws a parent by Thompson sampling on the steering counts of nodes that have already been evaluated, and it expands the child that parent proposes. Every method then returns the earliest evaluated node whose steering-split solve count is maximal. The tables report the held-out score of that node on the other 100 HumanEval+ instances and the other 145 Polyglot instances, next to the root scaffold and the calibration edit with the highest fit-split label. Twelve runs of GPT-4o return a node at 87.0% on held-out HumanEval+, against 84.2% for the parent-rate ablation. On that split Huxley-G\"odel Machine selection returns 85.1%, the root is at 81.2%, and the calibration-best edit is at 82.8%. On the same split Claude Sonnet 4.5 returns 86.6%, beside 86.5% for Huxley-G\"odel Machine selection, 85.3% for the parent-rate ablation, and 85.1% for the calibration-best edit.

arXiv:2610.29110v1 Announce Type: new Abstract: Merging different models into a single model is highly desirable for multi-task deployment, particularly in regulated domains like healthcare. In such settings, models are often locally fine-tuned due to strict data privacy policies, necessitating that aggregation operates exclusively on the fine-tuned weights. Existing methods apply a uniform merge across all tasks, ignoring the geometric relationships among their learned subspaces. When task updates occupy misaligned subspaces, projecting them onto a shared low-rank basis discards task-relevant directions, a failure mode named \emph{subspace interference}. Through this study, we show that the per-task projection error of a merged model is governed by the principal-angle alignment between each task and the shared subspace, so that grouping geometrically compatible tasks provably reduces this error within each group. Motivated by this observation, we propose \textbf{Group-wise Rank-Aware Model Merging (GRAM)}, which uses the principal-angle geometry of task vectors to decide which tasks should be merged together and then performs a standard SVD-based merge within each coalition; the pipeline requires only the fine-tuned weights at merge time. With only one additional model ($\nK{=}2$), GRAM significantly mitigates subspace interference and consistently outperforms prior merging methods across language understanding (GLUE with Flan-T5), 8-task CLIP vision, and medical instruction tuning (Qwen2-7B-Instruct with four English/Chinese medical LoRAs).

arXiv:2610.29092v1 Announce Type: new Abstract: Multilingual large language models are increasingly deployed in settings where visible reasoning traces support verification, education, and safety auditing, yet standard evaluations typically measure only final-answer agreement or holistic response similarity. As a result, they establish outcome consistency without determining whether users across languages receive comparable supporting evidence. We introduce a trace-level divergence framework that compares visible reasoning traces as structured sequences, decomposing agreement into structural similarity, order consistency, and holistic similarity, with pair-specific calibration to enable comparison across languages. Across three reasoning benchmarks, seven languages, and six frontier LLMs, the natural baseline---embedding each whole response and comparing the two by cosine---systematically overstates agreement: $52.1\%$--$67.3\%$ of examples in the top quartile of holistic similarity are not in the top quartile of structural similarity. This dissociation is stable across sentence encoders and step segmentations, and blind bilingual annotation studies in two English-anchored language pairs show that human judgments align with structural similarity rather than with the holistic baseline. A same-language control shows that the level of divergence is language-specific: two English traces for paraphrases of the same question agree more closely than an English trace and its counterparts in other languages. Our results show that answer-level and holistic evaluations can substantially overestimate the consistency of multilingual explanatory artifacts, motivating trace-level evaluation for reliable multilingual deployment.

arXiv:2610.29091v1 Announce Type: new Abstract: Objective.To distinguish the contributions of acquisition context and photoplethysmography (PPG) to session-calibrated blood pressure (BP) tracking.Approach. The original comparison used three repeats of five-fold participant-level cross-validation in 653 AuroraBP participants, excluding calibration-labelled prediction targets. Matched analyses separated context and PPG and removed elapsed-time terms. Later exploratory follow-ups retained PPG, accelerometer signals and three initial cuff readings. Separate studies evaluated pulse-rate fusion and an intraoperative VitalDB model. Main results.The original wearable-state model achieved participant-level mean absolute errors (MAEs) of 7.697/6.139 mmHg for systolic/diastolic BP (SBP/DBP). Added context reduced MAE by 0.436/0.126 mmHg (97.5% intervals 0.305-0.573/0.053-0.200). In the matched factorial without the external PPG prior, PPG improved on calibration, time and protocol-provided posture/activity by 0.214/0.135 mmHg; removing time left the DBP interval including zero. Standard PPG descriptors contributed, but additional nonlinear/attractor benefit was not established. In a separate follow-up series, a sensor-history ensemble achieved 7.299/5.445 mmHg versus its static-rich reference's 7.363/5.470, small gains of 0.064/0.025 mmHg. Larger DBP errors did not improve uniformly. Quality-weighted fusion reduced pulse-rate MAE from 3.985 to 3.690 beats per minute. VitalDB MAEs remained 15.375/8.833 mmHg despite small PPG gains. Significance. The contribution is a calibrated evaluation framework with explicit controls, reproducible feature-cache-to-results analyses and retained negative findings. The sensor-history gain is exploratory after repeated dataset reuse. Independently estimated wearable states, complete raw-signal reproduction of the supplied richer features, external transfer and clinical adequacy remain unestablished.

arXiv:2610.29057v1 Announce Type: new Abstract: Users increasingly rely on large language models (LLMs) to analyse collections of independent inputs, such as product reviews or social media posts, and return a single aggregated answer. This setting requires multi-instance processing (MIP), where a model combines information from multiple instances into one prediction. We study a task in which the model must infer a binary label for each instance, and identify the most frequent label among the full instance set. The correct answer should be invariant to instance order. By evaluating fifteen LLMs across five datasets and seven instance order patterns, with each instance set held fixed across patterns, we find substantial order sensitivity. Accuracy can differ by about 90 percentage points between the best and worst order patterns, with models generally more accurate when majority-label instances appear at the start than when they appear in the middle or at the end. For subsets of models, available reasoning traces and gradient-based attribution reveal uneven emphasis across positions. Traces disproportionately mention early instances, while attribution scores are generally higher near the beginning and end than in the middle. These results show that instance order can systematically bias aggregated predictions, highlighting the need to evaluate MIP across different input arrangements.

arXiv:2610.29046v1 Announce Type: new Abstract: Large language models (LLMs) have shown considerable promise in synthesizing formal specifications from source code. However, existing approaches primarily focus on functional contracts and loop invariants, while largely neglecting frame conditions, which specify the memory locations that may or may not be modified during execution and constitute an essential part of formal specifications. To address this limitation, we propose an LLM-based approach for synthesizing ACSL frame specifications alongside functional contracts from C code. Our approach analyzes code property graphs (CPGs) to extract frame-relevant information from the input code and leverages this information to guide the LLM in generating accurate frame conditions. The resulting specifications capture both region-separation constraints and data dependencies that govern variable modifications. Experimental results demonstrate the effectiveness of our approach. On a benchmark of 84 ACSL-compliant functions from non-trivial real-world systems, our method successfully synthesizes error-free frame specifications for 92.9% of the functions, representing a 179% relative improvement over state-of-the-art techniques. On a second benchmark consisting of 51 functions, the success rate increases from 80.4% to 96.1%. Furthermore, when integrated as an auxiliary component into existing LLM-based ACSL-synthesis frameworks, our approach improves frame-specification generation performance by up to 61.9%.

arXiv:2610.29036v1 Announce Type: new Abstract: Explaining a time-series model requires both locating the evidence and extracting meaningful knowledge from it. Yet a single timestep is not a semantic unit: the meaningful patterns live across many samples, not in any one of them. Per-timestep saliency is therefore noisy, high-dimensional, and unstable. Existing methods address this by introducing fixed-size windows and treating them as independent players in a cooperative game. Evaluating a coalition of such players requires defining a ``baseline'' region, which prior work tackles with fixed counterfactuals (e.g., a flat line, noise, or a spectral surrogate). This choice frequently dominates the resulting explanation, obscuring the model's true decisions. To address these issues, we propose GRAIL, a region-based time-series explanation framework that uses domain-specific intervals, such as the P-wave and the QRS complex of an electrocardiogram (ECG) beat, as players. We describe each region with human-readable features and generate counterfactual baselines with an on-manifold conditional generator. Such a choice of players also admits non-temporal features: static covariates enter as singleton players attributed with the temporal regions on one scale, so a patient's tabular evidence (e.g., age, sex) is weighed directly against the signal. We attribute the predictions to the features via Owen values with common-random-number variance reduction. Experimental results on both synthetic and real data show that the proposed framework produces faithful and robust explanations, evaluated under matched on- and off-manifold operators, offering meaningful knowledge to domain experts.

arXiv:2610.29023v1 Announce Type: new Abstract: Large language model (LLM) agents are often improved by iterating until a fixed verifier is done. Once that verifier saturates, the loop has nothing left to climb. We instead score agents against the current population, and a frozen panel records whether anyone got better. The panel mean can rise inside the seeds by elimination alone. We therefore count a run as training only when its final mean beats the best generation-zero member. A merge is admitted only when the child moves toward the donor, past the child the same operator produces at zero dose, and the lower confidence bound on that movement reaches half the requested dose. That is the admission test. A detectable shortfall rejects the operator, and the test is otherwise undetermined. We keep realised fitness \phi as the selector and reference performance \theta as the unused ruler. Every measurement uses gemini-3.1-flash-lite on one negotiation task, and no fixed-oracle arm was run. In the open loop, tactic append grafts short donor rules and is admitted. Free rewrite moves only a third of the requested dose and is rejected. Template append pastes raw donor replies into the child. Its movement clears the bound on the first probe but not on the re-probe, and it is left undetermined. A held-out re-probe on the five situations the merge never saw confirms that both appends move, and it does not rescue free rewrite. Neither append rises with the dose. Constrained rewrite is admitted in the open loop and was not run in the closed loop, which changes only the operator. The population, the scores and the task stay fixed. No free-rewrite run beats the best generation-zero member. Every tactic-append run and every template-append run does. Sampled loop children keep that split. Free rewrite moves no further toward the donor, and both appends still clear the bound, by less than in the open loop.

arXiv:2610.29014v1 Announce Type: new Abstract: Coding agents are rapidly advancing from single-file generation to long-horizon repository-scale engineering, yet the evaluation of their capabilities to author interactive extended reality (XR) applications remains an open challenge. Unlike traditional software, XR applications operate within dynamic 3D physical environments where runtime robustness depends jointly on continuous user motion, spatial input, and environmental sensing. We introduce GenXR-Bench, the first benchmark for the end-to-end evaluation of coding agents in XR software development. GenXR-Bench contains 112 expert-authored tasks spanning 24 feature implementations, 21 bug repairs, and 67 application generation tasks. Held-out tests combine scripted and adaptive interactions with programmatic and visual assertions to evaluate core functionality and runtime behavior. GenXR-Bench's testing suite achieves 96.5% accuracy against blinded expert judgments of 124 candidate applications executed on consumer XR hardware. Evaluating 10 models on GenXR-Bench, we find that the strongest configuration achieves only 50.9% overall Pass@1. Failure analysis identifies recurring misunderstandings in asynchronous sensor lifecycles, physical grounding, and embodied interaction semantics. GenXR-Bench establishes a reproducible testbed for evaluating coding agents in the interactive and physically situated conditions of spatial computing.

arXiv:2610.29010v1 Announce Type: new Abstract: Radiation treatment planning requires balancing competing objectives to satisfy clinical criteria while accommodating physicians' preferences. This study aims to develop an evidence-driven, multi-agent compound AI system that integrates DRL-based automated treatment planning with LLM-based conversational plan refinement for proton PBS therapy. A policy network trained using PPO first generates multiple plans without human intervention, from which one is selected as the target plan and the others serve as references for subsequent refinements. For challenging cases requiring further adjustments, an LLM-based multi-agent system enables plan refinement through natural language interactions. The system comprises a planning agent that recommends evidence-supported adjustments, a chatbot providing a unified conversational interface, and an evaluation agent that retrieves clinical metrics and identifies constraint violations. To enhance domain-specific reasoning, the planning agent based on Qwen3-14B is designed to justify its recommendations using evidence from previously generated plans and post-trained on task-specific data using SFT and GRPO. The framework was trained and evaluated using 21 patients with H&N cancers, with 5 PPO-generated plans per patient. Of these, 50 plans were selected for instruction-guided refinement. Compared with the base LLM, the post-trained Qwen3-14B demonstrated improved instruction-following, information retrieval and domain-specific reasoning capabilities. The resulting adjustments more effectively improved user-prioritized objectives while limiting deterioration in competing objectives, yielding more clinically appropriate trade-offs in patient-specific plan refinement. The proposed compound AI system provides a user-friendly planning platform, offering a promising approach to personalized proton PBS treatment planning.

arXiv:2610.28989v1 Announce Type: new Abstract: Customized agent deployments define policies over tools, resources, and execution conditions. Providing these policies in the task agent's prompt does not reliably prevent violations. Training a guard requires examples of tool calls that comply with or violate those policies. We introduce AgentPolicyForge, a framework that uses executable environments to construct policy benchmarks and iteratively train policy guards. With this framework, we construct PolicyGym, a dataset across ten policy families using 698 scenarios for training data generation and 254 separate held-out scenarios for human-reviewed test sets. Tool calls are executed and their outcomes recorded to help determine whether they comply with the policy. Every violating case uses a permitted tool, requiring checks of arguments, resources, and context rather than tool identity alone. The case construction process supports iterative guard training: an attacker generates requests using retrieved strategies, and the guard is fine-tuned on balanced, environment-verified examples. Training improves policy enforcement even with a fixed archive. Updating the archive further reduces in-domain attack success compared with keeping it fixed. We compare policy-aware task agent self-regulation with a trained policy guard. On PolicyGym, the final guard-based configuration achieves a 40-44% relative reduction in attack success rate and fewer benign rejections than policy prompting alone.

arXiv:2610.28987v1 Announce Type: new Abstract: Persistent memory is a largely undefended surface for injection and poisoning, and the same risk limits memory sharing among agents. We present PerMIT, a memory architecture with provable integrity guarantees under an explicit trust assumption: only user-typed inputs and values confirmed at a gate are trusted. PerMIT mediates the release and use of untrusted environment-derived values through a finite memory interface, while preserving agent-driven multi-turn planning and tool choice. Over this finite interface, we prove two integrity theorems that are also machine-checked in Lean 4, which together ensure that environment-derived memory reaches a sensitive action only if confirmed at the gate and influences the agent's behavior only through what the gate displays. Every attack in our threat model thereby reduces to a single auditable decision, which a human, an LLM verifier, or both can make in practice. On Trojan Hippo, PerMIT records zero observed attack success with self-verification and lightweight verifiers. Because each confirmation is a logged verdict on exact content, endorsements accumulate across users, and sufficiently endorsed items become community-confirmed with no further verification required. Under explicit assumptions, we prove and corroborate in simulation a co-scaling property: as the community grows, the risks of promoting poison and of wrongly evicting benign knowledge fall, while per-user verification effort does not rise. On $\tau^2$-bench, communities of up to 90 autonomous agents use PerMIT to build shared knowledge bases with no direct memory injection promoted, and the resulting knowledge significantly improves new agents on held-out tasks, suggesting a path toward agents that safely self-improve at test time.

arXiv:2610.28982v1 Announce Type: new Abstract: While large language models are increasingly deployed as tool-calling agents, their safety evaluations still focus primarily on their ability to refuse harmful requests in text. We investigate whether the refusal behavior in text transfers to tool use. We construct a benchmark of 2,304 prompts across four domains, consisting of matched triplets that present the same harmful or benign request in text-only, standard tool-enabled, and adversarially framed tool-enabled settings. We then assess tool-call safety using 20 predefined forbidden actions. Testing five models of varying sizes and families, we find that requests refused in text frequently result in unsafe tool calls. Under standard tool-enabled prompting, this divergence occurs for 10-52 percent of harmful requests refused in chat. To mechanistically investigate this gap, we identify a refusal-associated direction in the residual stream of each of the five models. We find that in tool-enabled contexts, the activations along this direction decrease in four of five models. Furthermore, lower activation predicts unsafe tool calls across all five models (AUC 0.76-0.86). However, restoring the signal to the respective text-only value through activation patching only prevents a subset of unsafe calls, primarily by suppressing tool use rather than by producing safer calls. Steering along the refusal direction nearly eliminates unsafe calls only at strengths that also suppress benign tool use or degrade the output quality. Our findings indicate that the identified refusal direction contributes to whether a model acts, but only partially explains the refusal divergence. Therefore, evaluating the safety of tool-calling agents requires directly examining their actions along with their textual responses.

arXiv:2610.28945v1 Announce Type: new Abstract: Large language model (LLM) agents are increasingly expected to interact with environments and improve their decisions during deployment. However, how anticipating future outcomes and drawing on accumulated interaction experience can jointly support decision-making remains underexplored. We propose a test-time framework, inspired by Model Predictive Path Integral (MPPI) control, that combines multi-branch predictive analysis with experience-based adaptation. At each step, the agent proposes multiple candidate actions and imagines their outcomes with a world model, so that alternatives can be compared without being executed in the environment. Since language environments provide neither explicit dynamics nor costs, the predicted outcomes are evaluated for task progress and risk, and MPPI-style weights over the resulting costs guide the final decision. A dual-branch experience memory enables self-improvement: the planner branch refines candidate generation and action selection, while the world model branch improves next-observation prediction. Experiments on ALFWorld and ScienceWorld across three backbones demonstrate the strong overall performance of our method and the effectiveness of experience-guided predictive control. Our code will be made publicly available.

arXiv:2610.28934v1 Announce Type: new Abstract: Coding agents now complete tasks autonomously, modifying real-world state such as artifacts, software environments, and configurations along the way. Yet they show limited backup awareness: recognizing that an action will modify state that may later be needed, and preserving it beforehand. To evaluate backup awareness, we introduce RegretBench, a benchmark of 205 executable tasks grounded in regret moments reported on platforms such as X.com, GitHub, and Stack Overflow, where users explicitly regretted not preserving state they later needed. Across six frontier models, agents complete 86.1% of tasks on average yet back up before these moments in only 15.2%, showing that task completion ability does not imply backup awareness. Because it is hard to predict which changes will be regretted, we propose BackUpper, a harness plug-in backup layer that checks every tool call and preserves affected state before modification. It combines rule-based strategies for common operations with a lightweight subagent for complex cases, and its rule table can expand with subagent decisions over time. On RegretBench, BackUpper covers every regret moment action and restores states with 92.2% average similarity to the original. Achieving this costs only $0.014 per task, with 7x fewer tokens than the main-agent self-backup baseline. When users have preferences on what to back up, BackUpper can also narrow its scope accordingly, matching human annotations on whether an action is worth preserving with an 84.5% F1 score. These results suggest that agents can keep actions reversible at low cost while respecting users' preferences.

arXiv:2610.28914v1 Announce Type: new Abstract: Frontier language-model agents excel at complex tasks, but high inference costs make transferring their capabilities to smaller models essential. Traditional knowledge distillation achieves this through parameter updates, requiring access to the student's weights and additional training compute while encoding the transferred knowledge implicitly in the parameters. We introduce D-Skill, a framework that treats the teacher as a black box and distills its behavioral advantage into an interpretable natural-language skill by analyzing paired rollouts and identifying behavioral divergences and agreements. The resulting lessons are consolidated offline, with any compatible skill-update backend, into a skill document that is provided to the student at inference time. D-Skill constructs the skill entirely offline from fixed trajectories, without labels, further environment interaction, or weight updates after rollout collection. Across SpreadsheetBench-Verified, ALFWorld, and chess puzzles spanning multiple teacher-student pairs, D-Skill closes 75%-95% of the teacher-student gap without touching model weights. After a single offline pass over the paired traces, D-Skill outperforms one-epoch SFT baselines and remains competitive with multi-epoch SFT. Under the same model family, the resulting student also incurs 5.7x-8.3x lower inference cost per task than the teacher. Our results demonstrate that natural-language skill distillation offers a performant, interpretable, and efficient alternative to weight-based capability transfer.

arXiv:2610.28912v1 Announce Type: new Abstract: This paper charts the evolution of the Prolog textbook Simply Logical -- Intelligent Reasoning by Example from print with runnable programmes on a 3.5-inch diskette published in 1994, via various intermediate online versions, to the fully interactive online edition available today. Our aim is to investigate -- from both the writer's and the reader's perspectives -- the potential impact of technology on educational materials. The three authors of this paper present three distinct and complementary points of view, which come together to shape an interactive online book that offers an immersive learning experience. Peter describes the philosophy behind the original book and experiences teaching from it. Jan demonstrates how contemporary web programming has enabled a fully interactive realisation of the book's philosophy. Kacper reports how different technologies can be adopted to build a versatile authoring toolkit that underpins the interactive online edition of the book. Collating these distinct yet coherent perspectives allows us to reflect on future opportunities for interactive Prolog teaching materials arising from the continuous development of web technologies. The insights we offer should be of interest to teachers of Prolog as well as those wanting to find out about the latest educational technologies.

arXiv:2610.28909v1 Announce Type: new Abstract: Deep research systems iteratively plan, gather evidence, synthesize, and revise, making end-to-end latency a key bottleneck for interactive use. Almost all production systems are built on autoregressive (AR) language models and inherit left-to-right decoding in planning and long-form writing. We present DiffResearch, an open multi-agent deep research framework that places a diffusion language model (dLLM) at the core of every stage, and measure how a diffusion backbone changes latency and report quality. Holding the agentic scaffold, retrieval code, prompts, and evaluation fixed, we swap the backbone between an open diffusion model and a comparable AR model. Across 100 Deep Research Bench tasks, the diffusion backbone is roughly 6 times faster end-to-end, driven by order-of-magnitude reductions in planning and writing. We further introduce dFast, a decomposition mode that reads intermediate denoising states and extracts subqueries before generation finishes, providing a 1.99 times median speedup for decomposition in isolation with no loss of report quality. On the benchmark's RACE rubric, dFast obtains the best overall score among the three configurations. Finally, in a public system-level leaderboard snapshot, an instance built on a stronger diffusion model ranks above several commercial systems. We release DiffResearch under an Apache-2.0 license.