跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+367新增论文
+288其中 AI 相关
+33已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 10:18 · 全站最近入库：2026-11-09 10:17各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 10:48 — 2026-11-09 10:48（北京时间） · 点击数字查看对应简报；AI 相关包含在新增论文内。

最近完成检查：2026-11-09 10:18最近发现新增版本：2026-11-09 10:17库内论文最新更新时间：2026-11-06 12:44

每分钟刷新已保存的数据。空检查不代表采集异常；新收录可能包含历史论文。AI 导读按新增版本计数。

17,467 已收录论文14,949 AI 相关4,222 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

时间范围跟随排序：版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常，查看状态 →

找到 3,417 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 10:16（北京时间）
新论文 · v1arXiv:2610.37316

ReForest: Interpretability Meets Reasoning in Isolation Forest

Juyoung Seo, Geonwoo Bang, Dongho Kim, Moohong Min

原文摘要

Statistical anomaly detection methods like Isolation Forest are efficient but suffer from a critical semantic gap. They identify numerical rarities but fail to understand the contextual meaning of data, often missing subtle, context-dependent anomalies. Conversely, Large Language Models possess strong semantic reasoning capabilities but are prohibitively expensive and slow for large-scale deployment. To bridge this gap, we propose ReForest, a novel framework that synergizes the statistical efficiency of Isolation Forest with the reasoning power of LLMs. ReForest transforms the latent decision paths of Isolation Forest into natural-language explanations, enabling an LLM to refine anomaly detection through an uncertainty-aware dynamic ensemble. This approach allows the model to adaptively weigh statistical and semantic signals based on uncertainty-driven instance-level confidence. Extensive experiments on four real-world datasets (Healthcare, Finance) demonstrate that ReForest significantly outperforms baselines, achieving an average 21% improvement in Recall and a 10% gain in F1-score. Our framework not only improves detection performance but also enhances model transparency by grounding semantic narratives in statistical evidence, providing interpretable rationales essential for high-stakes decision-making.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2503.13383

An Embarrassingly Simple Data Selection Strategy for Million-Scale Multi-Modal Instruction-Tuning

Mengyao Lyu, Yan Li, Huasong Zhong, Hui Chen, Jungong Han, Guiguang Ding, Zhenheng Yang

原文摘要

Recent research demonstrates that pretrained large language models (LLMs) require minimal but carefully curated data for instruction-tuning. Building on LLMs, multi-modal models further contend with far larger data volumes, heterogeneous sources, and cross-modal alignment, amplifying both the necessity and the complexity of data selection. However, existing methods rely on deeply extracted embeddings, gradients or narrow quality scores that are time-consuming to compute and limited in expressiveness, frequently failing to surpass a simple random-sampling baseline. To overcome these limitations, we step back from complex embedding spaces and target the fundamental objective of instruction tuning: how and what knowledge should be elicited from the pretrained model and exposed to users. We leverage interaction \underline{S}tyles as a proxy for structural diversity, and decompose data quality into capability-specific \underline{S}cores, yielding an explicit yet \underline{R}ich (SSR) profile for each sample. We further cast the selection as submodular coverage maximization over the capability--style grid, providing an approximation guarantee. SSR consistently delivers superior performance across two distinct architecture stacks, LLaVA-1.5 (Vicuna + CLIP) and LLaVA-OneVision (Qwen2 + SigLIP), two large-scale data pools, six data budget settings, and 14 multi-modal benchmarks. To our knowledge, SSR is the first data selection method validated at million-scale, achieving 99.1% of full-data performance using only 30% of a 2.6M pool. Furthermore, the SSR profile models and selected subsets transfer training-free across data pools and model architectures with capability extensibility.

cs.CV · cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37314

A Lightweight Attention-Based CNN with Saliency-Map Interpretability for Skin Cancer Detection

Sanjana Islam Kasfia, Md. Ismiel Hossen Abir, AKM Bahalul Haque, Abir Mahmud Shahariar

原文摘要

Skin cancer is among the most prevalent and life threatening forms of cancer worldwide. Early and accurate diagnosis significantly improves patient survival rates. However, manual dermoscopic assessment remains challenging due to visual similarity. This study proposes an Explainable AI-driven framework for automated binary classification of skin lesions using a lightweight attention-based convolutional neural network (CNN-Attention). The model integrates channel attention to focus on informative regions of dermoscopic images while maintaining low computational complexity. A publicly available dataset comprising 2,637 training and 660 test images was used, with balanced class weights. The proposed model achieved 87 percent accuracy with 1.95 million parameters and 7.45 MB of storage, matching the best baseline (MobileNet) while using approximately half as many parameters and half as much storage. ROC AUC and PR AUC scores reached 0.9475 and 0.9430, respectively. Additionally, saliency maps were incorporated as an explainability technique to visualize regions that influence model decisions. Comparative analysis with VGG16,VGG19,ResNet101,and MobileNet demonstrates the proposed models effectiveness in balancing classification results, computational efficiency, and interpretability.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2505.22843

On the Reliability and Stability of Selective Methods in Malware Classification Tasks

Alexander Herzog, Aliai Eusebi, Lorenzo Cavallaro

原文摘要

The performance figures of modern drift-adaptive malware classifiers appear promising, but does this translate to genuine operational reliability? The standard evaluation paradigm primarily focuses on baseline performance metrics, neglecting confidence-error alignment and operational stability. While prior works established the importance of temporal evaluation and introduced selective classification in malware classification tasks, we take a complementary direction by investigating whether malware classifiers maintain reliable and stable confidence estimates under distribution shifts and exploring the tensions between scientific advancement and practical impacts when they do not. We propose Aurora, a framework to evaluate malware classifiers based on their confidence quality and operational resilience. Aurora subjects the confidence profile of a given model to verification to assess the reliability of its estimates. Unreliable confidence estimates erode operational trust, waste valuable annotation budgets on non-informative samples for active learning, and leave error-prone instances undetected in selective classification. Aurora is further complemented by a set of metrics designed to go beyond point-in-time performance, striving towards a more holistic assessment of operational stability throughout temporal evaluation periods. The fragility we observe in SOTA frameworks across datasets of varying drift severity suggests it may be time to revisit the underlying assumptions.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37281

RPSD: Retrieval-Privileged On-Policy Self-Distillation with Optimal Transport

Dongyue Li, Zebang Cheng, Zhongyi Deng, Hongboxu, Yi Bin, Fei Ma

原文摘要

Sequential recommendation predicts a user's next interaction from historical behavior. Recent LLM-based methods recast this task as generative recommendation and increasingly adopt reinforcement learning such as GRPO for post-training. However, these methods suffer from two limitations, namely credit-blind optimization, where a sequence-level reward is shared across all generated tokens, and collaborative-blind supervision, where each training signal is derived from a single user and ignores behavioral patterns shared across similar users. To address both issues, we propose Retrieval-Privileged on-policy Self-Distillation (RPSD). For credit-blind optimization, we introduce on-policy self-distillation (OPSD), in which a teacher with privileged knowledge re-evaluates the student's on-policy response, and the token-level confidence difference between the student and teacher reshapes the shared sequence-level advantage into token-specific credit. However, the privileged knowledge in vanilla OPSD is still derived from a single user, leaving collaborative-blind supervision unresolved. We therefore ground the teacher in collaborative behavioral evidence from similar users. Specifically, we use temporally aware optimal transport (OT) to compare interaction trajectories across users by jointly considering item similarity and temporal structure, and further apply diversity-aware selection to retain complementary trajectories as privileged evidence. In this way, on-policy self-distillation provides fine-grained token-level credit, while OT enriches the teacher with structured cross-user evidence. To the best of our knowledge, this is the first work to bring on-policy self-distillation to sequential recommendation and to use cross-user behavior as its privileged signal. Extensive experiments across multiple real-world datasets demonstrate that RPSD achieves state-of-the-art performance.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37276

Where Should the Evaluation Budget Go in LLM-Driven Agent Scaffold Evolution?

Sichun Luo, Yi Huang, Mukai Li, Haochen Luo, Guanzhi Deng, Haibo Wang, Jian Xu, Zefa Hu 等 10 位作者

原文摘要

Large language model (LLM)-driven agent scaffold evolution improves agent workflows through iterative candidate generation and evaluation. Evaluation both guides subsequent search and determines the final scaffold, but evaluating every candidate on the full task set incurs substantial execution costs. We investigate how evaluation budgets can be allocated across these two roles. Our experiments show that sparse feedback can sustain productive evolution, while additional targeted evaluation improves final selection. Replay analyses further show that shared informative items and a common scoring reference support comparisons among closely matched candidates. Motivated by these findings, we propose \method, an evaluation allocation framework that combines sparse search-time feedback with targeted final confirmation. A short fully evaluated warm-up establishes a shared informative core and calibrates a response model whose item parameters remain fixed during search. After search, \method refits the model and applies paired successive halving, reusing accumulated observations and executing only missing evaluations. Across four benchmarks and two evolution frameworks, \method achieves higher mean performance than full evaluation in most settings while using approximately half as many evaluation cells.

cs.LG · cs.AI · cs.CL · cs.NE

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2605.06993

Optimal Experiments for Partial Causal Effect Identification

Tobias Maringgele, Jalal Etesami

原文摘要

Causal queries are often only partially identifiable from observational data, and experiments that could tighten the resulting bounds are typically costly. We study the problem of selecting, prior to observing experimental outcomes, a cost-constrained subset of experiments that maximally tightens bounds on a target query. We formalize this as the max-potency problem, where epistemic potency measures the worst-case reduction in bound width guaranteed by an experiment, and show that this problem is NP-hard via a reduction from 0-1 knapsack. Building on the polynomial-programming framework of Duarte et al. (2023), we give a general procedure for evaluating epistemic potency in discrete settings. To control the super-exponential search space, we introduce two graphical pruning criteria that depend only on the causal graph and the query: a novel path-interception rule that exploits district structure to certify zero potency in linear time, and an identifiability check based on the ID algorithm. On Erdos-Renyi random graphs and 11 bnlearn benchmark networks, the two criteria together prune 50-88% of candidate experiments on average without solving a single polynomial program. For the general subset search, we show that ID-pruned experiments are combinatorially inert, yielding a super-exponential reduction in the number of subsets evaluated. We close with an end-to-end demonstration on observational NHANES data, selecting optimal experiments for estimating the effect of physical activity on diabetes.

cs.AI · stat.ML

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2610.16925

Comparing GenAI to Machine Learning and Manual Review for eDiscovery

Gordon V. Cormack, Maura R. Grossman

原文摘要

Keeling et al. [Keeling2026Generative] report on a case study recently conducted by Redgrave LLP in which they assert that Relativity's Generative AI Review tool (aiR) achieved 88% recall while Relativity Assisted Learning (RAL), its machine-learning tool, achieved 64% recall. These results are not comparable because the methods employed have different objectives, because of inter-assessor disagreement, and because they represent different recall-precision and cost-benefit trade-offs. Our analysis of the Redgrave data indicates that, when these differences are taken into account, aiR and manual review are comparable. Furthermore, established machine-learning models employing Continuous Active Learning (CAL) and Scalable Continuous Active Learning (S-CAL) show state-of-the-art effectiveness, and superior efficiency, when trained only on a sample of 2,000 first-pass-reviewer responsiveness determinations.

cs.LG · cs.AI · cs.IR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37245

Aligning Large Language Models for Safety in African Languages: A Case Study of Hausa

Valdini Douglace Lemofouet, Blessing Ngozi Uzor, Paula Chikaodinaka Anyanwu, Danielle Blanche Kapsa, Saminu Mohammad Aliyu, Seid Muhie Yimam, Shamsuddeen Hassan Muhammad

原文摘要

Safety alignment research for large language models (LLMs) remains concentrated on high-resource languages, leaving open how existing alignment methods perform in low-resource, culturally distinct settings. We study this question for Hausa, a language spoken by more than 200 million people across West Africa. We introduce HausaSafeLM, one of the first Hausa safety preference dataset, built from 10,000 prompts derived from PKU-SafeRLHF: 3,000 human-corrected (gold) and 7,000 machine-translated (silver) examples spanning six harm macro-categories. Using Qwen2.5-7B-Instruct, we compare supervised fine-tuning (SFT) and SFT followed by Direct Preference Optimization (DPO) under both data regimes, evaluating on 500 harmful and 200 benign Hausa prompts.We find that the untuned base model appears highly safe by refusal-rate metrics alone, but this reflects limited Hausa generation ability rather than genuine alignment: its refusals are almost uniformly low quality. SFT improves both refusal fluency and precision and benefits substantially from the larger, noisier silver-tier data, while DPO sharpens refusal precision but comes at a steep, largely data-invariant cost to linguistic quality. No single configuration jointly maximizes safety, helpfulness, and fluency. Our results suggest that current alignment techniques teach models when to refuse more readily than how to do so fluently in underrepresented languages, and that refusal-rate metrics alone are an unreliable signal of safety in low-resource settings.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37243

TCM-DiagCommit: A TCM-SD-Derived Benchmark for Evidence-Conditioned Syndrome Selection and Commitment

Erwei Hou, Yue Hu

原文摘要

Traditional Chinese Medicine (TCM) syndrome differentiation is usually evaluated as static final-label selection over complete clinical records. This setting hides three diagnostic behaviors: whether a model distinguishes semantic-neighbor syndrome labels from easy random distractors, whether it updates predictions as evidence fields are revealed, and whether it should commit under uncertainty. We introduce TCM-DiagCommit, a benchmark derived from the public TCM-SD/ZY-BERT resource while retaining its source syndrome labels; it converts 1,834 source cases into 23,842 task rows over 148 syndrome labels. The benchmark combines random, profile-neighbor, and full-vocabulary candidate ladders with staged t1/t2/t3 evidence. It also includes fixed-label tongue/pulse perturbations and insufficient-evidence variants. A benchmark-defined action contract scores commit, acquire, defer, and abstain. Across five zero-shot DirectLLM calibrations, random10 top-1 accuracy is 63.2% to 66.4%, but hard20 falls to 24.7% to 30.3% and full148 to 14.8% to 22.3%; on hard20, WT(all) wrong-terminal commits remain 62.2% to 74.3%. These results show that final-answer accuracy alone is insufficient for diagnostic benchmark evaluation: answer selection, evidence use, and benchmark commitment scoring should be measured separately.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37242

Grounding LLM Agents in Business Processes: Process-Aware Knowledge Graphs for Reliable Multi-Step Service Invocation

Saiful Khan, Afrah Farea

原文摘要

Large language model (LLM) agents that call enterprise APIs must select the correct endpoints, follow process order, use valid payload keys, and choose coded values from predefined lists. However, current retrieval-augmented service discovery treats endpoints independently and therefore misses process and schema constraints. We address this limitation by integrating Open Data Protocol (OData) metadata and Business Process Model and Notation (BPMN) reference processes in a typed property graph. A hybrid retriever identifies candidate endpoints using dense embeddings and follows process edges to retrieve subsequent endpoints. We evaluate the method on 42 enterprise services with process-chain ground truth. For queries with low lexical overlap between the query and downstream entities, it improves Recall@3 by 5.7 points (n=74), with consistent gains across embedding models and cut-offs. It matches dense retrieval on other queries and provides no improvement on OpenAPI services without a process layer, indicating that the gains arise from process structure. We also evaluate an LLM agent on 81 multi-step queries. It completes 38.7% of tasks, compared with 10.3% for a chunk-based baseline. In an elicitation probe, providing valid code lists eliminates invented codes but leaves invalid keys in 46% of write runs. Full typed context reduces invalid calls to 3% (3 of 101 runs), with no invented codes. These results show that exposing process and schema constraints during retrieval improves the reliability of agent-driven API calls.

cs.AI · cs.SE

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37239

When Is a Relation Worth Trusting? Manipulation Costs, Exposure, and the Trust Trap in Multi-Relational Graph Learning

Vincess Dongmo, Astride Melvin Fokam Ninyim

原文摘要

Multi-relational graph detectors learn from clean data how much to trust each relation, but an adversary pays a different price to manipulate each relation and each action; taking such costs as given, we ask when using a relation is worse than ignoring it. Our first main result proves that degree gating strictly beats every selection or reweighting of a relation when part of the population is isolated on it and the attacker cannot afford to open the gate, and that this window is sharp; on real graphs, validated gating gains exactly on the relations that meet these conditions and nowhere else. Our second shows that a per-node exposure bound, which extends to message passing, tells a platform which relation to harden: exposure per unit cost, computed without any attack, picks the relation whose hardening raises the exact trap threshold most in 15 of 16 cases, against at most 7 for informativeness, cheapness, or an analysis with equal edit costs. Supporting tools give exact worst cases for linear relational models, certified brackets for R-GCN and tree ensembles, and a finite-sample guarantee for adversarial validation, under which clean-trained graph detectors fall below their graph-free counterparts in 58 of 90 settings on six graphs. A selection protocol built on these results is never significantly worse than ignoring the graph; with at most four relations it mostly abstains, as validation alone would, and its cost-aware proposals help when relations are many, so far shown on synthetic graphs. Without cost information, the safe choice is to abstain.

cs.LG · cs.AI · cs.CR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37237

GraphRAGProv: GraphRAG Meets How-Provenance

Wadie El Amrani, Luis Galárraga

原文摘要

GraphRAG systems often use knowledge graphs to retrieve relevant context for an LLM that produces the final answer. That makes answers more inscrutable and therefore less appealing for domains where transparency is paramount. We present GraphRAGProv, an architecture that executes SPARQL queries over source-grounded RDF graphs to compute answers and their how-provenance polynomials. These polynomials record which statement occurrences were used and how they were combined. Because each occurrence is linked to its source text, users can follow these links to check whether the extracted statements reflect the text and whether the query combines them to answer the question. Querying thereby becomes a way to inspect and evaluate the text-to-graph transformation. Our interactive demo examines 400 cases from two multi-hop QA benchmarks, revealing extraction and query errors even when the final answer is correct.

cs.AI · cs.DB

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37224

Spine: Compiling Subtask Trees into Structured Context for Long-Horizon Coding Agents

Jiahong Xiang, Kunqiu Chen, Lingming Zhang, Yuqun Zhang

原文摘要

Long-horizon coding tasks contain nested subtasks. Typically, agents record interactions sequentially into a growing history, causing context inflation and performance degradation. Compressing or trimming the history without explicit subtask boundaries can discard information needed by unfinished subtasks. Conversely, recursive subagent delegation isolates completed subtasks through child agents by assigning every subtask boundary to delegation, which conflates subtask granularity with executing agent choice, adding execution and coordination overhead. We introduce Spine, an interface and runtime that align model context with subtask nesting and completion. An agent declares explicit subtask boundaries through work scopes, allowing the same agent to execute nested subtasks inline while reserving delegation for independent parallel tasks. The runtime organizes execution in SpineTree, a persistent tree recording subtask relations, local interactions, completion states, and returned results. Before each model call, it compiles the tree into model context just in time, retaining unfinished subtasks' interactions and replacing completed subtasks' details with their results, while preserving prefix-cache reuse. We integrate Spine into Codex to build SpineCodex. With GPT-5.6 Sol, compared with native Codex, SpineCodex improves SWE-Milestone Score by 11.8% while reducing total cost by 32.4%, improves the average ProgramBench test-pass rate by 17.3%, and improves Mean@5 on eight of the ten evaluated FrontierSWE tasks. On SWE-Milestone, it also achieves a comparable Score to Recursive Subagent at 73.7% lower total cost. We also integrate Spine into Pi through a reusable software package: with Gemini 3.8 Flash and Qwen 3.8 27B, it improves SWE-Milestone Score by 9.5-17.5% and the average ProgramBench test-pass rate by 4.3-8.7% compared with native Pi.

cs.SE · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2607.19358

LISA: Linear-Indexed Sparse Attention for Efficient Long-Context Reasoning

Yu Zhao, Zekun Zhang, Fan Jiang, Bo Zeng, Xintong Wang, Chonghan Liu, Shimin Shan, Yichao Du 等 9 位作者

原文摘要

Long chain-of-thought (CoT) reasoning models such as DeepSeek-R1 keep lengthening their inference chains under test-time scaling, yet standard self-attention incurs quadratic cost in context length, making long-CoT deployment costly in production. We therefore introduce LISA (Linear-Indexed Sparse Attention), a plug-and-play replacement for self-attention that requires no full pretraining. LISA combines a linear-attention branch for long-range memory with a Lightning Indexer that dynamically selects the most relevant tokens for a sparse self-attention branch. The two branches are fused through a learned gate, so the model retains broad context while focusing computation on informative tokens. We train LISA in two stages. In Stage 1, a linear-attention branch is integrated with sliding-window attention and distilled from a frozen teacher's full-attention distribution. In Stage 2, the Lightning Indexer replaces the static window and is trained with a per-head KL-divergence objective to mimic teacher attention patterns. The resulting model reduces eager-mode self-attention computation from quadratic in sequence length to linear in the number of generated tokens and the selected context budget. Experiments on DeepSeek-distilled Qwen models show that LISA cuts measured eager-mode mean sample latency by 23.9% at 16K context without degrading long-context reasoning quality. Across nine mathematical and code reasoning benchmarks, the gated LISA-1.5B model improves the equal-weight benchmark average by 4.2 percentage points relative to the full-attention baseline.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37215

Retrieve to Act: Motion Primitive Graph Retrieval for Robust VLA Execution

Pham Tri Quang, Bao-Ngoc Dao, Minh Duc Nguyen, Duc-Duy Nguyen, Hung T. Ho, Linh Dang Le, Trong-Bao Ho, Thien-Loc Ha 等 18 位作者

原文摘要

Vision-Language-Action (VLA) policies demonstrate remarkable semantic generalization, yet remain highly vulnerable to out-of-distribution (OOD) visual shifts during execution. While existing test-time methods attempt to suppress spurious visual cues, they fundamentally overlook the rich, uncorrupted kinematic priors embedded in the policy's own training experience. We introduce Retrieve to Act (R2A), a test-time scaling framework that formulates OOD robustness purely as a geometric motion primitive retrieval problem. R2A bypasses fragile visual features by structuring successful training demonstrations into an offline spatiotemporal motion primitive graph. During inference, a query-dependent Graph Neural Network dynamically localizes the robot's ongoing execution within this graph, retrieving structurally compatible future motion primitives. These retrieved geometric priors are then seamlessly harmonized with the frozen VLA's predicted actions via a learned, dimension-wise gating mechanism, adaptively rectifying erroneous spatial trajectories while preserving the policy's local reactivity. Extensive experiments across three state-of-the-art VLA architectures, GR00T N1.7, $π_{0.5}$, and StableVLA, demonstrate that R2A consistently and significantly enhances robustness against diverse contextual perturbations. R2A achieves absolute success rate improvements of up to +12% in simulation benchmarks (LIBERO-Plus, RoboTwin2.0-Plus) and +40% in real-world deployments, without requiring any finetuning of the base policies. Project page: https://retrieve2act.github.io/r2a/

cs.RO · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.27895

The Shape of the DPO Hyperparameter Landscape: What to Measure, How It Moves with Scale, and When Not to Transfer It

Hannan Mahadik, David Salinas, Jacek Golebiowski, Aaron Klein

原文摘要

Direct preference optimisation (DPO) is widely used for language model post-training, but its performance depends strongly on the choice of its hyperparameters, such as the learning rate and the reference-policy regularisation parameter $β$. To better understand the DPO hyperparameter landscape, we conduct extensive experiments using a dense hyperparameter grid across different model scales from the same family. This allows us to analyse how the optimal DPO hyperparameters evolve with model size and the number of update steps, with intermediate checkpoints providing an inexpensive signal to accelerate the hyperparameter optimisation process. Furthermore, by evaluating our models on both static and dynamic benchmarks, we identify which benchmarks provide the strongest signal for hyperparameter selection. Finally, we investigate how different state-of-the-art hyperparameter optimisation strategies perform in this setting. Our experiments show that the optimal learning rate varies with model size and that hyperparameters are not transferable across training datasets. We also find that validation loss is a poor proxy for hyperparameter optimisation, as hyperparameters that minimize validation loss only weakly correlate with optimal hyperparameters for downstream benchmarks. Therefore, DPO hyperparameters need to be tuned specifically for the target model, dataset, and evaluation objective. We also show that short candidate training runs can substantially reduce the cost of this selection. We release all our training configurations and evaluation results to support reproducible analysis of selection quality and training cost.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37734

When Skills Tell Agents Not to Ask: A Large-Scale Study of Oversight Directives in Agent Skills

Timofey Sanko, Yuan Tian

原文摘要

Agent skills, reusable instruction packages centred on a SKILL.md file, extend AI agents on demand, and a single line in a widely copied skill reaches every agent that loads it. Some lines are oversight-loosening directives: they tell the agent to proceed without approval, to bypass checks with --no-verify, or to keep working without checking in. Although such directives can remove a checkpoint from every agent that loads them, no prior work has examined them. We present the first large-scale study of oversight-loosening directives, analyzing 1.53 million English skills from GitSkills. We split skills into units (sentences, bullets, table cells, and code lines), develop a codebook that labels oversight directives by polarity, mechanism, and scope, label 2,097 units by hand, and train a detector to scale the analysis. Our results show that about one in seventeen skills (about 88,000 skills) contains a loosening directive. These directives most often loosen oversight through permission flags, and most of them (70.1%) apply unconditionally rather than only in specific contexts such as CI. Compared with skills that have the same amount of oversight language, skills that run shell code, install packages, or use sudo have 1.7-1.8 times the odds of containing a loosening directive, an association that holds across all our robustness checks, and they do not compensate with more tightening. Loosening spreads through a few upstream skills rather than through edits by their adopters: 2.4% of the skill families that carry loosening account for half of all copies of loosening-carrying skills, and 1.1% of distinct loosening units for half of all loosening occurrences. Open-source skill scanners catch at most 46% of loosening directives, with a precision below 0.4, while the LLM behind them, prompted with our codebook instead, catches 61% at a precision of 0.71.

cs.CR · cs.AI · cs.SE

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37201

Does Your Generalist Skill Optimizer Really Incentivize Agentic Tasks? Rethinking Skill Optimization via Bipartite Attribution

Jiawei Zhou, Bo Peng, Yuxuan Huang, Ming Zhang, Youbang Sun, Xiang Wang, Chaochao Lu

原文摘要

Skill optimizers adapt an agent to long-horizon tasks, such as repairing a terminal environment, by rewriting its skill, a natural-language procedure in its context, and keeping the edits that raise success on a small dev set. On such tasks success is noisy, so a gain on a small dev set does not show that an edit helps, and a success score does not show what the edit costs. We introduce BiPA, which replays an optimizer's acceptance rule on candidates identical to its seed to measure what selection alone reports, and records each episode as a bipartite graph of turns and the text they read to trace extra input tokens to the behavior behind them. Across three optimizers, three agentic benchmarks and two models, the learned skills succeed about as often as their hand-written seed, 0.1 points higher on average. Their dev gains vanish on test: keeping the best of several candidates on at most 30 tasks reports gains this large even when none changed. They read about 16% more input tokens: the optimizers almost always add rules, half to re-check work and some to act one call at a time, and the agent follows them as extra turns, each re-sending the whole conversation. Four principles follow: accept an edit only if its gain exceeds what selection alone reports; charge it for the turns it adds, not for its length; let the editor delete and merge rules; show it what earlier rules cost. Code is available at https://github.com/valleysprings/BiPA.git.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37195

Action-Value Calibration via Continuation Sharing for Long-Horizon Agentic Reinforcement Learning

Yingjian Chen, Lang Zhou, Zixin Zhao, Pingfan Li, Weishi Zheng, Zhilin Zhao

原文摘要

Reinforcement learning for long-horizon agents relies on feedback from sequences of interactions, making credit assignment to individual actions difficult. Existing methods such as GiGPO compare actions at shared states to compute relative advantages, but estimate each action's value from a single sampled continuation. The resulting local credit remains sensitive to decisions made later in the trajectory. To address this problem, we propose ConCal (Continuation Calibration), a recursive continuation-value estimator that incorporates continuation evidence already present in a rollout batch. ConCal matches visits from the same task by their incoming state-action-state transitions, combines shared continuation evidence with each visit's own transition estimate, and propagates calibrated values backward to refine earlier action-value estimates. This reduces dependence on individual sampled continuations and mitigates local credit distortions without additional rollouts. Held-out return diagnostics show lower error in within-group value differences and higher agreement with reference action orderings than estimates based on individual continuations. With Qwen2.5-1.5B-Instruct, ConCal improves success rates over GiGPO by 3.39 and 5.86 percentage points on ALFWorld and WebShop, respectively. Our code is available at https://github.com/Yingjian-Chen/ConCal.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37733

Selective Transfer Learning for Groundwater Level Anomaly Forecasting: Accuracy-Risk Trade-offs

Yaxuan Wu, Ali Mohammed Mansoor Alsahag, Seyed Sahand Mohammadi Ziabari

原文摘要

This work studies context-aware selective transfer learning for multi-well groundwater anomaly forecasting in the Netherlands. Source-domain pretraining can raise average accuracy but is not uniformly reliable: it degrades performance on a substantial fraction of heterogeneous wells. Context-Aware Selective Learning (CASL) uses calibration-trained, constrained contextual blending of frozen no-transfer and fixed-transfer parents. Global Convex Blend uses the same parents with one calibration-fitted weight per seed. The corrected task requires nine consecutive observed monthly inputs and an exact three-calendar-month target, without interpolation. Evaluation covers 457 wells and 13,718 test windows under five predefined seeds, with the four-cell maxsim0.60 source pool frozen using Always-TL validation RMSE alone. Fixed transfer harms 18.38-51.86% of wells. CASL has 0.424-1.084% lower RMSE than No TL, partially rescuing 97.53-99.50% and fully rescuing 23.46-43.78% of fixed-harmed wells. Partial rescue can arise structurally from convex blending and does not establish contextual superiority. Global Convex Blend protects more wells in four seeds, while CASL has slightly lower pooled RMSE in four. Neither dominates: the results support an accuracy-risk trade-off on the fixed split, not rolling deployment validity.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2605.30188

CalArena: A Large-Scale Post-Hoc Calibration Benchmark

Eugène Berta, David Holzmüller, Francis Bach, Michael I. Jordan

原文摘要

Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated. Post-hoc calibration provides a simple and widely used solution, but the large number of proposed methods, combined with small-scale and inconsistent evaluations, makes it difficult to determine which approaches are truly effective in practice. We introduce a large-scale, standardized benchmark for post-hoc calibration, covering nearly 2000 experiments across tabular and computer vision tasks, including binary, multiclass, and large-scale classification settings. Our benchmark aggregates predictions from a diverse set of classical models, modern deep learning architectures, and foundation models, and provides unified, reproducible implementations of dozens of calibration methods within a common evaluation framework. We argue that Post-Hoc Improvement (PHI) in proper scoring rules offers a principled alternative to traditional calibration error estimators for comparing post-hoc methods, capturing both calibration quality and potential degradation to the model's predictive performance. Using this framework, we conduct the most comprehensive empirical study of post-hoc calibration to date. Our results reveal consistent patterns across domains: smooth calibration functions outperform binning-based approaches, dedicated multiclass methods are essential in high-dimensional settings, and generic machine learning models are not competitive without calibration-specific design. To facilitate future research, we release all data, code, and evaluation tools, providing a plug-and-play benchmark for developing and comparing calibration methods.

cs.LG · cs.AI · stat.ML

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.32499

Learning an Anchored Prompt Space for Continual Adaptation of Large Language Models

Rongguang Ye, Zhan Zhuang, Yichen Wu, Ming Tang, Kede Ma

原文摘要

Continually adapting large language models requires acquiring new knowledge while preserving previously learned capabilities. Although jointly adapting model parameters and task-specific soft prompts is promising, it has two key limitations: previously learned prompts may lose effectiveness as the model evolves, and relationships among prompts that could support cross-task transfer are not explicitly learned. We propose Learning an Anchored Prompt Space (LAPS), which preserves historical prompt effectiveness and learns relationships among task-specific soft prompts to facilitate positive transfer. Historical prompts are first aligned with the updated model through self-distillation. LAPS then constructs an anchored prompt space whose vertices correspond to learned task-specific soft prompts and whose intermediate geometry is shaped by learnable Bézier control prompts. Once this space is learned, LAPS identifies the best-performing prompt for each task on its validation set, allowing the selected prompt to draw on knowledge acquired from related tasks. Experiments on the TRACE benchmark across three Qwen3 model scales show that LAPS consistently outperforms existing distillation-based, prompt-based, and joint prompt-parameter adaptation methods, improving overall task performance while reducing forgetting.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v7arXiv:2508.20866

AVIATOR: Towards AI-Agentic Vulnerability Injection Workflow for High-Fidelity, Large-Scale Code Security Dataset

Amine Lbath, Massih-Reza Amini, Aurelien Delaitre, Vadim Okun

原文摘要

The increasing complexity of software systems and the sophistication of cyber-attacks have underscored the need for reliable automated software vulnerability detection. Data-driven approaches using deep learning models show promise but critically depend on the availability of large, accurately labeled datasets. Yet existing datasets either suffer from noisy labels, limited vulnerability coverage, or fail to reflect vulnerabilities as they occur in real-world software. This also limits large-scale benchmarking of such solutions. Automated vulnerability injection provides a way to address these limitations, but existing techniques remain limited in coverage, contextual fidelity, or injection success. In this paper, we present AVIATOR, the first AI-agentic vulnerability injection framework. AVIATOR decomposes vulnerability injection into a coordinated workflow of specialized AI agents, tool-based analysis, and iterative self-correction, explicitly mirroring expert reasoning. It integrates RAG and lightweight LoRA-based fine-tuning to produce realistic, category-specific vulnerabilities without relying on handcrafted patterns. Across three benchmarks, AVIATOR achieves high injection fidelity (91-95%) surpassing existing injection techniques in both accuracy and vulnerability coverage. When used for data augmentation to train deep learning-based vulnerability detection (DLVD) models, AVIATOR provides the strongest downstream gains in vulnerability detection. Across models and base datasets, AVIATOR improves average F1 scores by +22% over no augmentation, +25% over VGX, holding the prior best injection success rate, and +3% over VulScribeR, the prior state-of-the-art LLM-based injection model, with +7% higher recall and no precision loss. Its augmented data exhibits the lowest distributional distortion and scales efficiently with <2% syntax rejection at 4.3x lower cost than VulScribeR.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.30027

Synthetic Hospital: An Open, Verifiable, Physician-Validated Longitudinal EHR Benchmark

Christine Park, Valerie Chen, Tim Dettmers

原文摘要

Frontier language models are rarely used in clinical workflows because the realistic, longitudinal benchmarks needed to develop them are scarce. Real electronic health record (EHR) data cannot be openly shared due to privacy, ethics or data use issues and it does not contain verifiable ground truth since the chart records only reflect what clinicians documented. We introduce Synthetic Hospital, an open, fully synthetic, fact-grounded longitudinal EHR benchmark that resolves the open sharing and verifiable ground truth barriers. Built entirely from public medical-education material with no protected health information, it comprises 1,268 longitudinal patients and 5,602 encounters, where every diagnosis, finding, and temporal relation is grounded in standard ontologies (ICD-10-CM, SNOMED CT, LOINC) and with a complete provenance chain back to its source medical education material. Synthetic Hospital is served through a simulated hospital record system that mirrors real EHR infrastructure (standard interoperability APIs, role-based access and function-calling interface). In a blinded review, physicians distinguished its records from real patient charts at near-chance rates (53%). Across 10 frontier and open models, none approaches ceiling: the best model reconstructs a patient's longitudinal problem list with a severity-weighted F1 of 0.73, level with the mean of seven physicians on a matched subset but well below the best of them (0.89), and misses roughly half of clinically relevant findings when summarizing a chart. Overall, these results highlight that Synthetic Hospital is a difficult and realistic test of clinical AI performance.

cs.AI · cs.DB

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37175HF 社区推荐

MLE-World: Predicting Machine Learning Experiment Outcomes with a Learned World Model

Brian Ko, Han Wang, Hejia Geng, Li Erran Li, Sherry Yang

原文摘要

AI research agents can propose, execute, and improve programs that solve machine learning tasks. However, executing each candidate program requires training and evaluating ML models, limiting how many candidates an agent can explore at each step. We introduce MLE-World, a world model finetuned from Qwen3.5-4B that predicts a program's terminal output (error traces, training logs, and validation scores) from its code and data description without executing it, together with a confidence score estimating the prediction's correctness. Because a prediction costs one inference call rather than a training run, MLE-World makes test time scaling (TTS) affordable for these agents. We build five TTS policies, guided by MLE-World, that compare several candidates per step by their predicted outputs: Sequential revision, Best-of-N, Pairwise judging, Fusion, and a UCB style rule that also adds an exploration bonus from this confidence score. Since predictions can be wrong, we run a candidate for real only when its confidence score is low. On 17 MLE-bench Lite tasks, MLE-World matches the real terminal output on 77.9% of held out programs, versus 41.1% for much larger prompted models. MLE-World guided TTS achieves up to 22.1% higher observed mean performance than the agent without a world model, with less real execution. Project webpage: https://mle-world.github.io/.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.16600

The Geometry of Learning: Read-Write Asymmetry Across Training

Valeria Ruscio, Keiran Thompson

原文摘要

Pretraining and alignment update the same transformer weights, and we ask in which directions each stage changes them. The residual stream gives every matrix a shared coordinate system, in which a weight change can be compared with the principal subspace of the activations the matrix reads and with the prediction subspace of the unembedding. In this frame, alignment updates to the query and key matrices are enriched in the principal input subspace, while its updates to the output matrices are close to random relative to the prediction subspace, and the same signature appears in late pretraining and under every objective we apply to one checkpoint, so it is a property of the training stage rather than of the objective. The gradient of a linear map is a sum of outer products $δ_t a_t^\top$, and the residual stream enters this product through the activation $a_t$ for matrices that read from it and through the upstream gradient $δ_t$ for matrices that write into it. Read gradients therefore inherit the spectrum of the activations, which is the same under every objective, while write gradients inherit the geometry of the upstream gradients, which point into the prediction subspace only early in pretraining, when the output matrices acquire their prediction geometry. The raw read gradient follows the activation spectrum, the optimiser attenuates the concentration it inherits, and cancellation between tokens sets the scale of the update but not its direction. A write direction derived from the unembedding still changes predictions after alignment, and adaptation restricted to any single group of matrices remains effective, so the signature describes where training places its updates, not which matrices can carry them.

cs.LG · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37169

Rènyi-Regularized Flow-Based Variational Inference for Bayesian Calibration

Exaucè Luweh Adjim Ngarti, Élise Arnaud, Léo Nicoletti, Victor Trappler, Arthur Vidard

原文摘要

Variational inference with normalizing flows provides a flexible posterior approximation for Bayesian model calibration, but struggles when the posterior has several modes separated by regions of low probability mass. We study this behavior in a two-parameter twin experiment and regularize the variational objective. Under Gaussian observation errors and a Gaussian prior, the corresponding minimization objective is the expected Tikhonov calibration objective minus the differential entropy of the approximating distribution. However, the expected Tikhonov term favors concentration near one of the modes, while entropy maximization does not guarantee multimodality. We introduce a regularization term based on Rényi divergence that encourages the approximate posterior to cover the prior support. In the connected bimodal case, a flow approximation represents both modes; in the case with well-separated modes, the flow and Gaussian mixture frequently miss one mode. The introduced regularization reduces the near-collapse fraction from 50% to 25%. While the regularization improves mode coverage in this experiment, it changes the Bayesian variational objective, introduces a trade-off in approximation accuracy, and does not provide a general solution to mode collapse.

cs.AI · stat.ML

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.34717

VLA-HWBench: Where the Time Goes When VLA Models Run on Edge Hardware

Borui Xu, Hongshi Tan, Wenqi Jiang, Bingsheng He, Ming Wu, Jason Zeng, Michael Heinrich

原文摘要

Vision-language-action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control. However, it remains unclear how to design VLA models, their inference software and edge hardware so that robots can run VLA inference in real time. Most VLA models report only task success, while prior studies cover limited action-generation strategies, divide inference into at most three stages, and lack comparisons of inference optimizations on the same models. Therefore, we introduce VLA-HWBench, a benchmark that measures 18 open VLA models on NVIDIA Jetson AGX Thor and AMD Ryzen AI Max+ PRO 395, with a seven-stage inference breakdown. Our study reveals three insights. First, action-generation strategy matters more than model size. It dominates 14 of 19 Jetson configurations, latency varies by $82\times$ on Jetson and $157\times$ on AMD, and parameter count explains at most 15% of latency variance. Models that generate long action chunks in a few flow-matching steps keep up with control, whereas token-by-token decoding does not. Second, kernel fusion with 4-bit weights helps most, running $π_{0.5}$ and GR00T N1.7 2.6-2.9$\times$ faster than PyTorch with torch.compile, but it exists for only these two models and only on the Jetson. Optimization also moves the time to the stages it shortens least, as CPU preprocessing grows from 12% to 39% of the GR00T N1.7 call. Third, after optimization, memory bandwidth limits only the stages that process few tokens on the Jetson, such as the denoising steps, whereas software limits all other stages, from unfused kernels to serial preprocessing on the host CPU. Overall, low-latency edge VLAs should use few action-generation iterations, inference software should extend kernel fusion and low-bit weights to more models and platforms, and edge hardware needs high memory bandwidth and a fast host CPU.

cs.RO · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37166

LiveSearch: Questions as Executable Verifier Code for a Live Benchmark of Professional-Source Financial Search Agents

Sinuo Wang, Wentao Zhao, Jiacheng Lu, Rui Sun, Beidi Luan, Qiming Shi, Jing Li, Daxin Jiang 等 9 位作者

原文摘要

In finance, the answer that matters is usually the current one: prices, rates, and disclosures hold only as of a given day, so retrieving up-to-date facts is a core capability of a financial agent. Yet most financial benchmarks overlook timeliness. They fix each answer at annotation time, so the gold goes stale as markets move and, once leaked into training data, rewards recall rather than search and analysis. We introduce LiveSearch, a live benchmark that stores each question as executable verifier code: a typed program of calls to a professional financial data service, followed by a declarative extractor. The question is frozen, while its gold answer is determined by executing the program against professional financial data at evaluation time, and a daily job re-executes it with no human or language model in the loop. A memorized answer is thus likely outdated, and a model must reach the current one through tools and analysis. Construction is equally automatic: a language model only verbalizes programs enumerated from real tool responses and never chooses an answer. Under general web search, six contemporary tool-calling models reach 70.9% to 76.7% answer accuracy. Over three consecutive days, 70% of served questions change their gold value, and even under tolerance-aware matching, a literally reused earlier snapshot fails on a majority of questions within a single day. Per-question tracking uncovers dynamic shifts in correctness masked by aggregate accuracy.

cs.AI

查看论文
arXiv ↗PDF ↗