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

+1072新增论文
+845其中 AI 相关
+94已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 14:26 · 全站最近入库：2026-11-09 13:22各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 14:51 — 2026-11-09 14:51（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 14:26最近发现新增版本：2026-11-09 13:22库内论文最新更新时间：2026-11-07 02:59

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。

18,172 已收录论文15,506 AI 相关4,527 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 5,189 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 14:25（北京时间）
版本更新 · v8arXiv:2609.38627

Marking Contour Tones in Yorùbá: A Typographic and Computational Proposal

Kólá Túbòsún

原文摘要

Yorùbá is a tonal language in which contour tones pose persistent orthographic challenges. These are especially notable for personal names and lexical items whose conventional spellings avoid vowel lengthening that would otherwise provide a host syllable for the second tone. A particular concern is a class of names in which the conventional spelling does not just omit tonal information but inverts the meaning of said name, sometimes asserting the opposite of what the name intends. This paper describes the problem, illustrates the inadequacy of current solutions, and proposes the adoption of the caron and circumflex marks. These are symbols with precedent in Yorùbá phonological scholarship since Olmsted (1951), used as orthographic conventions on single vowels to encode rising and falling contour tones, making them accessible for the first time through standard keyboard input and computational text processing. The proposal is supported by an implementation in the WriteYoruba keyboard and the TTSYoruba speech synthesizer, whose architecture and listener evaluation are reported separately (Tubosun, Oluokun, et al. 2026).

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02155

Decision Models as Policy Gates for Coding Agents: Explicit Policy Boundaries Improve Detection and Reduce False Alarms

James P. Beno

原文摘要

Coding agents can destroy data, leak secrets and message the wrong person faster than a human can review. Could a decision model catch these risks at runtime? We evaluated 22 hosted decision models and two prompted general models as classifiers for a multi-rule policy engine. A synthetic dataset of 2,643 tool calls was labeled blind against 28 rules, with a sealed holdout. With a single shared preamble in both conditions, explicit boundaries written as examples and counterexamples raised recall by 5.5-26.1 percentage points on all 23 available models and reduced false positives on 21. Jev 1.14's recall rose from 89.3% to 98.2%, while unnecessary interruptions fell from 30.6% to 4.8% of 252 clean actions. With examples, its median latency was 120 ms at $0.46 per thousand requests. Strong models retained the recall gains when concrete names were replaced by generic descriptions; three largely retained them when the same boundaries were written as prose. On 16 new scenarios written after the initial studies, a follow-up improved recall on 22 of 23 models and reduced unnecessary interruptions on 21. We release the dataset, rule texts and harness.

cs.CL · cs.CR · cs.SE

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02153

Multi-Agent Collusion and the J-Space

Marco Molinari, Gabriele Bettineschi

原文摘要

LLM agents are increasingly deployed in Multi-Agent Systems (MAS), and they introduce the risk of coordination among agents that might escape human oversight. The J-space has been used to uncover deception in single LLMs, but its use for uncovering and steering collusion between agents remains unexplored. We show, with interpretability experiments on DeepSeek-V4-Flash (284B) on the NARCBench collusion benchmark, that colluding agents' J-space holds the collusion itself: before replying, 86% of colluders represent misrepresentation, concealment or remorse. In a steganographic card-counting task, the receiver's J-space holds a hidden count, and causal interventions alter its count. We also introduce the Jacobian Pair Contrast (J-PACT), an unsupervised method for detecting collusion between agents. We evaluate J-PACT on four open-weight models: GPT-OSS-20B, Llama-3.3-70B-IT, Qwen3-32B, and DeepSeek-V4-Flash (284B). J-PACT retains 85% of the above-chance AUROC of supervised NARCBench probes, matches them on steganographic scenarios, and names the colluding pair up to its complement in 99% of detected runs.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02152

Detailed Derivation of The Backpropagation Algorithm

Faris Alasmary

原文摘要

Backpropagation is the standard method for computing the gradients that train neural networks, but it is usually presented in a compact vector form that skips the intermediate steps. This tutorial derives backpropagation step by step for a fully connected feedforward network with sigmoid activations and a squared-error loss. Using a running example with three layers, we first differentiate the error with respect to a single weight by following its path through the network with the chain rule. We then show how summing over all paths leads to the error terms $δ$ and to the four general backpropagation equations in matrix form. Every term is annotated with its dimensions, so each equation can be checked for consistency. The derivation is then extended to $m$ training examples, where the weight gradients become a single matrix product and the bias term corresponds to broadcasting. It is also extended to mini-batch gradient descent, including why the gradient of a mini-batch estimates the full gradient. Batch and mini-batch training are both given as algorithms. Finally, each equation is linked to the corresponding line of a NumPy implementation trained on the AHDBase dataset of handwritten Arabic digits, and the complete code is included.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.02372

PredActBench: Benchmarking Tool-Augmented Dialogue under Controlled Tool Noise

Abdulrahman AlRabah, Xiaocheng Yang, Dilek Hakkani-Tür, Abdussalam Alawini

原文摘要

Large language model (LLM) agents are increasingly used to support decisions by calling external tools. Existing work on LLM tool-calling assumes these tools are perfectly reliable. However, in many applications, the tools are statistical predictors with non-trivial errors. We introduce PredActBench, a benchmark for evaluating dialogue agents that work with imperfect predictive tools, using education as a testbed. First, we build a benchmark in which LLMs work with noisy predictors to guide users' decisions. Second, we extend two prior metrics for trust calibration, Relative AI-Reliance and Relative Self-Reliance, to multi-turn dialogue. Third, we evaluate 13 LLMs on OULAD and our released PredAct-CS dataset of 60 courses with real final grades and synthetic weekly scores, alongside a human study involving 13 instructors and teaching assistants. We find that when tools are noisy, SOTA models rarely make predictor uncertainty visible to users, with our strongest baseline GPT-5.5 achieving an F1 of 57.2. As a result, users over-rely on incorrect predictions and unsupported claims, showing that task accuracy alone is not enough for AI decision support and agents must communicate when their tools may be wrong. We release PredActBench to support the development of LLM agents that help users calibrate their reliance.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.38850

Beyond Outcome Rewards: Post-Rationale Distribution Optimization for Selective Prediction in Reasoning Models

Zhimin Gao, Meibo Hu, Jiamian Wang, Zhiqiang Tao, Pichao Wang

原文摘要

Reasoning models are trained to produce correct answers, yet selective deployment requires ranking which answers to accept or defer, and outcome-reward training does not directly optimize the post-rationale distribution used for that decision. We identify two mismatches: expected correctness is mode-seeking rather than distribution-eliciting, and scoring an average across rationales rewards disagreement unavailable to a single-rationale decision. We introduce Post-Rationale Distribution Optimization (PRDO), which applies a proper score to the answer distribution after each sampled rationale, without target rationales or separate confidence labels. Across four 0.6--3.8-billion-parameter models on MMLU-Pro, PRDO improves Brier score, error ranking and secondary accuracy relative to budget-matched, temperature-scaled GRPO. At 20\% realized error, oracle coverage rises by \npj{Q3a_vs_grpo.qwen3-0.6b.mp.oracle_cov@0.2.ppabs}--\npj{Q3a_vs_grpo.phi-4-mini.mp.oracle_cov@0.2.ppabs} percentage points, and ranking gains transfer to unseen tasks. Matched controls identify proper, per-rationale scoring as the active design choices, while verification tasks show the boundary: better selective ranking can coexist with lower full-coverage accuracy. Code is available at https://github.com/ZimmyGao/openjev-rlcd

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2606.05134

Activation-Based Active Learning for In-Context Learning: Challenges and Insights

Yaseen M. Osman, Geoff V. Merrett, Stuart E. Middleton

原文摘要

Deep active learning has previously been explored for LLM in-context sample selection, but not with methods that utilise recent advances in understanding of transformer activations. In this paper, we test the hypothesis that model activations could provide a fine-grained signal to optimise the selection of in-context examples. We present a comprehensive analysis of MLP activation-based deep active learning methods applied to in-context learning, including how different attention masking strategies impact active learning across diverse classification and generative datasets, using both Llama-3.2-3B and Qwen2.5-3B base models. However, we find a negative result: MLP and embedding layer outputs, viewed through the lenses of massive activations or the first four moments, do not correlate with example quality or task performance. Specifically, the absolute Spearman correlation coefficient is at most 0.33 for all tasks and models we tested, showing that such activation-based sampling should not be used for in-context learning. We hypothesise that this may be due to superposition, whereby models represent more features than they have dimensionality, suggesting that methods like Sparse Autoencoders (SAEs) may be a promising future direction.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02134

HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response Safety

Ashmiya Lenin, Navaneeth Sangameswaran, Preetham S

原文摘要

An input guard decides whether a request reaches the model, but never sees what the model says. Harm that emerges during generation, because a jailbreak passed the input layer or a benign request drifted into operational detail, is visible only on the output side and only as tokens arrive. We present HaloGuard 2.0, an open-weight streaming output guard that assigns a safe/unsafe probability to every response token in a single forward pass, so a deployment can stop the stream once harm is confirmed. The guard reuses the constitution of HaloGuard 1.0, our input guard, to organise a 478,532-response training corpus, and supervises each token from the point where the response becomes harmful. We find that a guard trained on conventionally shaped data learns a shortcut specific to streaming: it judges the prompt instead of the response, and handles only 13.9% of matched safe/unsafe response pairs correctly. We correct this with a second phase on 106,043 counterfactual responses built around the base guard's own failure modes, and release the equal-weight average of the base and corrective models. On six English response-safety benchmarks, HaloGuard 2.0-0.8B reaches a macro F1 of 83.6 (83.3 on held-out halves of the decision-rule search), up from 72.8, with 38% fewer false positives. It matches the strongest full-response guards with 5-10x fewer parameters and is 2.6 points above the strongest streaming guard, Qwen3Guard-Stream-4B. A multilingual variant, HaloGuard 2.0-Poly-0.8B, trained on PolyGuardMix, reaches 83.1 F1 on the response labels of PolyGuardPrompts across 17 languages, 5.0 points above the strongest baseline. On one H100 the guard costs 0.37 ms per token per stream when moderating 64 concurrent responses. Both guards are hardened continuously through the weekly red-teaming protocol of HaloGuard 1.0, extended here to the output layer.

cs.CL · cs.CR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02129

FOLLIE: Automated Reasoning over the First Dataset of Sentences with their First-Order Logical Representation for French

Maximos Skandalis

原文摘要

FOLLIE (for First-Order Logic for Language Inference and Entailment for French) connects French sentences, compositional first-order representations and automated natural language inference (NLI). The first French NLI-ready, NL-to-FOL resource, it contains 19,983 inference records and 36,987 non-empty sentence-slot--formula mappings from six heterogeneous NLI sources, retaining French lexical material and source problem structure. Its silver annotations are produced by a neural--symbolic pipeline, with 111 separately identified LLM-assisted additions. The construction combines morphosyntactic analysis, probability-conditioned categorial multitagging, grammatical proof search, discourse representation and first-order semantic construction. Stage-wise experiments show substantial gains in derivation coverage; analyses of conjunction distributions, logical-operator incidence and scope alternation characterise the resulting resource. We then develop an automatic inference engine whose input-conditioned background theories address lexical relations, quantification, quantities, event structure and temporal dependencies. Prover9 supplies proofs for entailment and contradiction, while Mace4 supports bounded consistency checks. Three-way accuracies are 65.37% on all 335 FraCaS-FR problems, 70.33% on 300 GQNLI-FR problems and 71.35% on 9,399 eligible SICK-FR records pooled across splits. Agreement with reference labels among committed decisions is 93.53%, 100.00% and 98.66%, at commitment coverage of 41.49%, 37.00% and 15.05%, respectively. Worked constructions, axiom schemas and proof analyses explain these outcomes and identify the contributions of scope, reference, lexical knowledge and numerical structure. FOLLIE provides a French experimental basis for studying the representations and assumptions that connect sentence interpretation to deductive inference.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02121

Conditional Calibration Contracts under Coupled Load: A Branch-Aware Audit of Switched Inference

Juan S. Santillana

原文摘要

Bursty serving changes both queue states and the population selected for each model. We study conditional calibration contracts as a specification to be validated, rather than a certified property of an existing router. CA-MHD combines load-coupled distillation, closed-loop replay and a cost-informed Schmitt router in a discrete-event simulator with a hardware service table. Across 40 cells (ten training seeds and four content-load coupling levels), we distinguish the original teacher, frozen wide heads on an adapted trunk, and the narrow student. A common teacher-trajectory audit has median absolute student-teacher deficit 0.06 percentage points, but does not certify the student-selected deployment support. Historical results use a kernel correctness gap, not the published SmoothECE; we separately prove selected-branch composition for a residual functional at a common fixed scale. The deployed wide branch has lower ordinal accuracy than the original teacher in all 40 cells. Same-checkpoint student-only serving has lower historical ordinal gap in 27/40 cells; routing also exposes violation tails up to 17.5 percent. Student p99 latency remains above the 1.5 ms target. Retrospective temporal recalibration improves the ordinal gap in 39/40 cells, but causal delayed-label availability is unverified and no incremental per-mode benefit is detected after multiplicity adjustment. An in-sample CUSUM detects 30/40 checkpoint changes at five historical false alarms, with no prospective false-alarm guarantee. These findings support a branch-aware audit and a scoped negative result, not demonstrated routing superiority. Independent budgets, selected-support validation and net utility under a satisfied SLA remain open.

cs.LG · cs.CL · cs.DC

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02112

Order-Only On-Policy Distillation

Guangzeng Han, Yu Wu, Xiaolei Huang

原文摘要

On-policy distillation (OPD) trains a student on its own generations by matching the teacher's token probabilities. These probabilities entangle two kinds of information: which tokens the teacher prefers, and how confident it is. The latter reflects the teacher's own capacity and calibration, which a smaller student may be unable to reproduce. Because the teacher temperature changes these values without changing the ordering, OPD is also sensitive to the temperature choice. We ask whether the teacher's ordering of candidate tokens is enough, and propose Order-Only On-Policy Distillation (O$^{3}$PD). At each student-generated prefix, O$^{3}$PD trains the student to reproduce the order of the teacher's top-$K$ tokens, without fitting their probability values. The objective is therefore invariant to the teacher temperature and needs no temperature tuning. Across six teacher--student pairs, O$^{3}$PD outperforms OPD on the average over HMMT25, AIME24, and AIME25 in every pair, by up to 2.9 points, with gains of up to 5.6 points on AIME25. Other ranking losses yield similar gains, so the benefit comes from ordinal supervision itself. Analysis shows that O$^{3}$PD departs further from the teacher's probabilities than OPD, yet reproduces its ranking more faithfully and yields a stronger student. The ranking, not the probability values, thus carries the knowledge that on-policy distillation needs to transfer.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02104

Consolidating Reasoning with Test-Time Training

Haotian Wu, Zeming Chen, Hao Zhao, Antoine Bosselut

原文摘要

Effective multi-agent collaboration requires finalizing solutions from reasoning that is distributed across the individual attempts of many agents. However, useful intermediate reasoning can remain unexploited when collaboration is limited to voting on conclusions reached among individual agents, or aggregating long, noisy collections of reasoning trajectories in context. In this work, we propose TTT-Consolidate, a framework that uses multi-agent reasoning trajectories as adaptation data to improve a model's subsequent reasoning. TTT-Consolidate jointly consolidates trajectories into parametric representations through test-time training, then generates a new solution by conditioning on the consolidated information. We meta-learn how to consolidate using bi-level optimization at training time, where an inner loop encodes reasoning trajectories into a lightweight LoRA adapter, and an outer loop learns meta-parameters for the inner loop such that the consolidation is helpful for the model to synthesize a final solution. As no large-scale corpus of diverse reasoning trajectories exists for training aggregation methods, we construct the OpenParallelThinking corpus composed of 12,810 math problems paired with a mixture of trajectories from six LLMs. Evaluated on challenging math and STEM benchmarks, TTT-Consolidate outperforms the best aggregation baselines by up to 20 percentage points, even under matched compute and runtime settings, with the gap widening at larger model scales. TTT-Consolidate also recovers from unsuccessful individual trajectories more robustly, an ability that improves as the number of available (even incorrect) trajectories increases. Finally, we demonstrate that TTT-Consolidate benefits from both SFT and RL-based outer loop objectives. We will release TTT-Consolidate and OpenParallelThinking for use in future work on reasoning aggregation.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.38960

Efficient Agentic Audio-Visual Question Answering through Toolset Pruning and Modality Consolidation

Yi Wang, Tianzi Wang, Peter Bell

原文摘要

Fine-grained audio-visual question answering (AVQA) requires models to resolve complex cross-modal correspondences and temporally grounding from sparsely distributed evidence. Omni-modal agents, which autonomously orchestrate perception tools to actively collect evidence, achieve strong performance, but it remains unclear whether active perception from both audio and visual modalities depends on broad specialized toolsets, especially audio or visual modality-specific tools. Such systems may also introduce long execution trajectories and substantial inference cost. We investigate whether comparable performance can be retained with a compact workflow exposing only essential perceptual capabilities. Starting from the specialized-tool OmniAgent, we conduct staged controlled experiments that prune the toolset and progressively consolidate modality-specific tools into omni-modal tools, motivating a two-tool agent comprising only Omni-modal captioning and Omni-modal QA. Across three widely used AVQA benchmarks, our compact agent achieves comparable or better accuracy than the specialized-tool agent while reducing inference cost by around 70%.

cs.AI · cs.CL · eess.AS

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.25421

Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making

Guangfeng Cai, Kaibing Yang, Shuo He, Yu Li, Shengtian Yang, Jiaqi Lv, Lei Feng

原文摘要

Recent studies on world modeling for Large Language Model (LLM) agents typically formulate the learning objective as next-observation prediction. However, this objective ties supervision to what a transition happens to reveal, which may omit the dynamics most relevant to the agent's current decision. To bridge this gap, we propose Agent-Authored World Modeling (AAWM), a training procedure that constructs supervision from the policy's own decision needs. Specifically, at each state, the agent identifies what it needs to understand about the environment before acting. These needs drive the retrieval of relevant transition evidence across trajectories, which is then synthesized into training targets that capture decision-oriented dynamics instead of reconstructing the next observation. This aligns the training objective with the dynamics the policy needs before acting, not with the contents of the next observation. Experimental results validate the effectiveness of AAWM across multiple environments and training settings. These results show that decision-aware world-model targets provide a more effective learning signal than next-observation prediction.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02079

ParetoSaddler: Pareto-Guided Optimization of Agent Harnesses for Quality and Cost

Xingzhi Qian, Sungho Park, Rongyuan Tan, Jianbo Yu, Jue Zhang, Lorenzo Cavallaro, Menglin Xia, Yanzhi Chen 等 16 位作者

原文摘要

Harness optimization can substantially improve LLM agents without modifying model weights, but existing methods primarily optimize response quality. Extending them to joint quality--cost optimization introduces a challenge: applying the deployment preference during intermediate updates can prematurely reject candidates that are valuable for later improvements. We introduce ParetoSaddler, which addresses this mismatch by separating intermediate candidate promotion from final deployment selection. At each iteration, an optimizer updates a parent harness to produce a candidate, and then compares the two over repeated executions and uses Bayesian bootstrap Pareto support to promote candidates showing credible progress in either quality or cost. Promoted candidates receive broader development evaluation and remain available as parents for later optimization steps. We evaluate ParetoSaddler on ClawMark and GAIA2 using domain-disjoint splits. On ClawMark, ParetoSaddler improves quality from 70.86\% to 81.35\% while reducing cost from \$1.17 to \$1.01 per task; on GAIA2, it improves quality from 56.22\% to 62.89\% while reducing cost from \$0.78 to \$0.45 per task. The Pareto-selected harnesses also achieve higher quality at lower cost than quality-only harness optimization. Analysis shows that Pareto-aware optimization retains intermediate candidates that would be discarded by deployment preference but can support later improvements, while Bayesian bootstrap screening filters out candidates whose gains are weakly supported across runs.

cs.AI · cs.CL · cs.LG · cs.MA

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02033

How Green Is Your Manifesto? Scoring Environmental Positions in the 2024 EU Elections with an LLM Panel

Franco Corti, Sajid Siraj

原文摘要

Locating political parties on environmental and climate policy is central to the study of party competition, yet the established instruments, expert surveys and the hand coding of manifestos, are costly, periodic, and slow to extend to new documents. We ask whether a panel of large language models can serve as a cheaper and reproducible alternative. Treating each manifesto as a set of sentences, three models from independent providers score every sentence on two dimensions, environmental protection and climate change, separating a party's position on an issue from the salience it gives it, and the judgments are averaged to the manifesto level. Agreement across the models is retained as a built-in reliability signal. We apply the panel to 41 manifestos from the 2024 European Parliament election, covering Austria, France, Germany, Italy, Spain, and the European party families, and validate it against the 2024 Chapel Hill Expert Survey for the 25 parties with a like-for-like expert benchmark, using the other manifestos as diagnostic cases. The panel recovers the expert ordering of parties with strong rank agreement on both dimensions, while reading them as systematically greener than experts do, a gap we interpret as the distance between manifesto promise and governing record. Inter-model disagreement modestly flags where the instrument errs, and position and salience prove related for environmental protection but independent for climate change. The result is a transparent, low-cost, and multilingual instrument that complements rather than replaces expert and hand-coded approaches; the prompt, code, and manifesto-level results are released openly.

cs.CL · cs.CY · econ.GN

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.31090

Poolwise: Evaluating Reward Models for Selection across Prompts

Mateusz Modrzejewski, Tomasz Radzikowski

原文摘要

Reward models are used to select training data, but common benchmarks compare responses only within the same prompt. Selecting data across prompts also depends on how scores compare between prompts. We introduce Poolwise, a protocol that selects a fixed fraction of a response pool by score, measures the selected set's quality using external labels, and reports within-prompt ranking separately. In matched experiments on two rating sets and two model sizes, training each prompt's mean score to match its average response rating improves selection over permuted or omitted mean targets in every seed at 30 percent retention. In grouped HelpSteer experiments, the model with the highest within-prompt accuracy selects a lower-rated set than the correct-mean model. Subtracting prompt means estimated from separate reference responses preserves local rankings but lowers selected helpfulness for six open reward models and two closed judges on HelpSteer. Poolwise provides a direct way to compare reward models and score transformations for data selection.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02021

ClickSight: An LLM Framework for Interpreting Student Clickstreams Through Learning Strategies

Bahar Radmehr, Lan Tran, Ekaterina Shved, Fatma Bet{ü}l G{ü}reş, Adish Singla, Tanja K{ä}ser

原文摘要

Digital learning environments produce rich clickstream data that can reveal how students approach learning tasks, yet interpreting these fine-grained traces often requires substantial expert effort. We introduce ClickSight, a framework for interpreting student clickstreams through learning strategies defined in the literature. Given a student clickstream and a predefined set of learning strategies, ClickSight uses a staged LLM-based pipeline to identify strategy-relevant evidence, assess adherence to each strategy, and generate an evidence-grounded interpretation of the student's behavior. We evaluate ClickSight across two learning environments, 365 students, and 12 model backbones using an expert-grounded quality rubric and an LLM judge calibrated with human-expert ratings. Our results show that ClickSight can generate high-quality strategy-based interpretations across different models and environments, with reasoning models performing more consistently than instruction-tuned counterparts. We further find that providing external clickstream-extracted guidance can improve interpretation quality for strategies that require precise quantitative information from the trace. By connecting fine-grained clickstreams with theoretically defined learning strategies and preserving the evidence underlying each judgment, ClickSight provides a flexible approach for studying student behavior in digital learning environments.

cs.AI · cs.CL · cs.CY

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02018

CULTURE-CMT: Benchmarking Contextualized Social Media Translation

Linjuan Wu, Dingming Li, Yiwen Qiu, Xinze Lyu, Ye Guo, Daoxin Zhang, Zhe Xu, Weiming Lu

原文摘要

Social-media comments are interactions, yet they are often translated and evaluated as isolated text. This mismatch is especially problematic when meaning depends on surrounding discussion, community-specific slang, or culturally situated expressions. We introduce CULTURE-CMT, a Chinese-English benchmark for contextualized social-media comment translation that restores the post and conversational context naturally available to users. Its 572 manually verified examples are organized into Slang, Context, and Both, enabling diagnostic analysis of distinct and interacting sources of translation difficulty. To support scalable evaluation, we further develop a retrieval-augmented cultural-effectiveness Judger that combines a calibrated scoring model with dynamically retrieved cultural knowledge. CULTURE-CMT reveals several non-trivial patterns. Providing interaction context improves most model configurations, but does not eliminate culturally ineffective translations. Examples requiring both contextual grounding and culturally situated language remain substantially more difficult than either source of difficulty alone. Moreover, larger models are not consistently better, and thinking produces highly model-dependent gains, including occasional degradation. In contrast, a 27B model adapted on benchmark-disjoint social-media data achieves the lowest observed invalidity rate among the evaluated configurations under thinking inference with context. Together, these results suggest that realistic social-media translation is not simply a matter of scaling models or reasoning longer, but of learning to translate meaning as it is constructed within evolving online interactions.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02017

Switch-SWE: Do Coding Agents Hold Focus When Switching Between Unfinished Tasks?

Shuochen Chang, Qianli Ma, Yibo Miao, Qingyang Liu, Bingjie Gao, Haonan Zhao, Siyu Sun, Yulin Sun 等 10 位作者

原文摘要

Software development sessions often involve several unfinished tasks rather than one issue carried through to completion. As coding agents take on repository-level development, they need to recover an earlier task's requirements and partial progress after intervening work. Existing repository-level benchmarks typically score one issue-to-patch episode and do not directly measure whether an agent can resume a task after advancing other unfinished work in the same repository. We introduce \textbf{\ourmethod}, a benchmark for focus-task resumption in sessions that interleave unfinished tasks drawn from the same source repository, with a persistent checkout for each task. Each focus task is evaluated in matched uninterrupted and switched sessions with identical focus requirements and executable verifiers. The switched session interleaves ordered stages from other unfinished tasks in the same repository, while only the final focus patch is scored. \ourmethod~contains 200 focus tasks from 19 Python repositories, reconstructed from issue, pull-request, and review histories and rendered as 400 matched session specifications. Across 7 coding agents, interleaved unfinished work produces an average focus-task resolution gap of 11.5 percentage points, ranging from 6.0 to 15.0 percentage points. Trajectory analysis identifies cross-task requirement confusion and loss of task state as recurrent resumption behaviors. We introduce a plug-in context-management module that selects task-relevant history and maintains concise task-state summaries. On OpenCode, the module improves focus-task completion under same-repository task switching by 5.5 to 15.5 percentage points across the evaluated models. Code is available at github.com/Miraclecsc/Switch-SWE.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02015

Scaling Makes It Possible: How Large Models Master Impossible Languages

Yaning Jia, Chunhui Zhang, Chiyu Ma, Zhongyu Ouyang, Ivory Yang, Soroush Vosoughi

原文摘要

While language models (LMs) excel in natural language tasks, their ability to learn impossible languages--those that violate the structural, cognitive, or logical constraints that make human languages learnable and functional--remains debatable. This controversy is further fueled by a divide between linguistic theorists and empirical researchers working with small-scale LMs. To address this controversy, we introduce semantic coherence. This concept measures how effectively semantic information is expressed in language structures and examines the learning capabilities of LMs across multiple scales. By fine-tuning billion-parameter LMs on impossible language samples, we demonstrate that larger LMs achieve similar performance on impossible and possible languages, whereas million-parameter LMs exhibit significant degradation when trained on impossible languages. This performance gap widens as semantic complexity increases. We further show that scaling up the LM size fundamentally shifts the learning focus from grammatical rules to semantic understanding, with billion-scale LMs exhibiting a threefold reduction in sensitivity to low-level linguistic perturbations while maintaining stable performance under high-level semantic perturbations.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02011

MedPsy: State-of-the-Art Small Medical Language Models for Efficient Edge Deployment

Davide Vitabile, Alexandro Buffa, Akshay Nambiar, Amril Nazir

原文摘要

Medical language models promise expert clinical support, but the strongest open-source LLMs remain too large for private edge-device deployment, while compact medical models still trail substantially on knowledge-intensive and real-world healthcare tasks. Reasoning models that fit edge memory budgets can produce long reasoning traces, increasing sequential decoding, end-to-end latency, energy consumption, and KV-cache memory pressure. We present MedPsy, a family of text-only 1.7B and 4B medical language models designed for edge deployment, targeting this accuracy-generation-efficiency trade-off. Our recipe combines a synthetic medical data pipeline over biology, medicine, and health seeds with chain-of-thought targets from a 235B medical-focused LLM teacher; a four-stage post-training curriculum comprising two SFT stages followed by two RL stages with hard-sample mining; and an edge deployment evaluation spanning multiple GGUF quantization variants and real-device inference on laptop and smartphone hardware. On seven closed-ended medical benchmarks, MedPsy-4B surpasses MedGemma-1.5-4B by 19.34 points and matches MedGemma-27B while being 6.75x smaller; MedPsy-1.7B outperforms MedGemma-1.5-4B by 11.42 points despite being less than half its size. On HealthBench-Hard, MedPsy-4B surpasses MedGemma-27B by 15.33 points, while MedPsy-1.7B surpasses it by 11.66 points despite being ~16x smaller. Beyond accuracy, MedPsy reduces average response length by 1.7x (1.7B) and 3.2x (4B) versus its Qwen3 backbones, while Q4_K_M quantization preserves the aggregate score within 1 point of BF16 with ~69% lower disk footprint. In real-device inference, MedPsy-4B's shorter generations translate into 75% lower median end-to-end latency and estimated energy per answer than its Qwen3 backbone at matched quantization, while peak RAM remains below 5 GB across our tested laptop and smartphone configurations.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02009

Retrieval-Enhanced Cultural Evaluation for Social Media Translation Training

Linjuan Wu, Yiwen Qiu, Xinze Lyu, Dingming Li, Hui Yao, Zhongyi Sang, Ye Guo, Daoxin Zhang 等 14 位作者

原文摘要

Social media translation requires preserving meaning while capturing the cultural nuances of user-generated content (UGC). However, gaps in the cultural knowledge of LLM-based evaluators can lead to inaccurate assessments and unreliable training rewards. To improve these signals, we use retrieval-enhanced cultural evaluation to train CUE-MT, a family of social media translation models. A calibrated, domain-specific Judger guides supervised data refinement and is augmented by an Agent for hard-example mining and reinforcement learning. The Agent retrieves relevant external evidence as needed and distills it into concise hints that the Judger uses to score translations and assign rewards. Across 9B, 27B, and 35B-A3B backbones, CUE-MT outperforms the evaluated open-source baselines of comparable size on CUE-MT-Bench and three public benchmarks. Human evaluation on CUE-MT-Bench shows cultural effectiveness gains of 7.44-15.12 percentage points over the strongest external baseline in each size group. Retrieval improves the Judger's four-way classification accuracy by 8.07 percentage points over a matched no-retrieval baseline that uses LLM-generated hints alone. Training ablations further support the combination of hard-example mining and Agent-enhanced rewards.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02002

To Align Long, Plan Short: Simulating Diverse Profile-Aligned Students by Realizing Short-Term Plans in Long-Horizon Learning Environments

Bahar Radmehr, Jiaxu Zhao, Fatma Betül Güreş, Adish Singla, Tanja Käser

原文摘要

Simulated students offer a scalable way to test learning platforms and AI tutors before deployment with real students and to augment scarce data for studying learning at scale. Existing LLM-based simulators primarily target short, language-centered interactions. In open-ended learning environments, however, student behavior emerges from patterns across long-horizon interaction trajectories, where turn-level prompting can drift from an assigned behavior profile. We introduce ALPS (Align Long-term by Planning Short-term), a hierarchical simulated student conditioned on observable, environment-specific behavior profiles. Rather than selecting every action independently, ALPS alternates between planning profile-aligned task segments and generating the corresponding actions. We evaluate ALPS in both data-sufficient and data-scarce settings, where supervision comes either from real student trajectories or from a teacher LLM. Across both settings, we compare in-context learning (ICL), supervised fine-tuning (SFT), reward-ranked fine-tuning (RAFT), and their sequential combination (SFT followed by RAFT). Experiments in an inquiry-based physics game and a pharmacy simulation evaluate behavioral density and coverage against held-out real data, and profile consistency using a human-aligned LLM judge. Across both settings, ALPS consistently outperforms turn-level generation, achieving comparable density and judge pass rates under scarce supervision, although coverage remains lower. These results show that hierarchical planning enables faithful, profile-aligned simulation over long interaction horizons, even with limited real student data.

cs.AI · cs.CL · cs.CY

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02001

TrackMem: Tracking Evolving Beliefs in Long-Horizon Financial Dialogue

Austin M. Y. Cheung, Harsh Vardhan Gupta, Go Nam Lui, Lok Kan Chan, Ginny Y. Wong, Simon See

原文摘要

Memory systems for LLM agents are evaluated mostly on social dialogue, where facts accumulate and rarely change, so remembering reduces largely to finding the turn in which something was said. Financial dialogue breaks this assumption: figures are revised, stances shift without being restated, rules stated once govern later questions, and a single company is discussed under several analytical lenses. Memory systems that retrieve the passages most similar to a question, or the most recent ones, therefore return superseded values, miss rules that share no wording with the question, and cannot recover a stance spread across sessions. We present TrackMem, a per-speaker graph memory that tracks the history of figures, separates perspectives by entity and lens, and preserves each speaker's actions in chronological order. To evaluate it, we introduce FinLoCoMo, a benchmark of 15 long-horizon investment conversations with 765 questions in two splits: one following the LoCoMo categories and one targeting four failure modes. With two readers, TrackMem outperforms existing memory systems on both splits. On the failure-mode split it reaches 73.9% macro accuracy against 69.4% for the strongest baseline, with the largest gains on questions about revised figures (86.5% against 73.4%).

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.30168

Finitely Verifiable Segmentation: Decidability for Every Tokenizer with a Finite-State Verifier

Nicklas Nidhögg

原文摘要

A segmentation policy assigns to each byte string at most one segmentation into tokens. Call it finitely verifiable when a finite automaton over marked strings, bytes paired with boundary bits, accepts exactly the policy's own segmentations. The automaton judges a proposed segmentation and need not compute one. From a supplied verifier the paper gives constructive decisions for four questions, whether a window certifies a token boundary, whether the boundary gap is bounded, whether a mask can keep generation from dead-ending, and whether two policies are equivalent. The last three specialize limitedness, co-accessibility and the equivalence of single-valued transducers. The first refines the synchronizing cut of code theory, and two more ask whether a certified cut, or the cuts of an inventory, let each piece scan alone. The class sits strictly between the online-scannable policies and those with regular domains, and online scannability is decidable in polynomial time through Choffrut's theorem. Markings are length-preserving, so the class is exactly the rational segmentation policies, a normal form rather than a restriction. The class is closed under four pipeline operations, composition over a labeled stage, restriction to a regular schema, splitting at character-class changes, and byte-fallback union of literal policies. Each composite inherits the six decisions, and certificates transfer between stages under four laws. Under rational, unifilar, almost surely stopping finite-state sources, each mass of the weighted section is an exact rational, and so is each rate where expected length is positive. Maximal-munch lexing, canonical byte-pair encoding and two-mode WordPiece instantiate the interface, and the munch library implements all six decisions. Every theorem carries an executable check, and every reported statistic is a program's own output.

cs.FL · cs.CL · cs.PL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.01967

Better Answers, Weaker Evidence Control: Evidence Bypass in On-Policy Distillation

Ruoran Xu, Yilun Zhao, Ziyu Chen, Xiaoqiang kang, Arman Cohan

原文摘要

On-policy distillation (OPD) can improve answer accuracy while weakening responsiveness to supporting evidence. We investigate this separation in scientific-document question answering through matched document interventions that change the supported answer while preserving the question and unrelated context. We identify evidence bypass: a student retains an originally correct answer after the document changes to support a different one. On QASPER with a Qwen3-4B student, standard OPD raises original-document accuracy by 3.4 percentage points, lowers counterfactual accuracy by 5.3 points, and increases bypass by 15.7 points; the same directional pattern appears on PeerQA. Controlled perturbations localize the shift to answer-critical evidence, and bypass grows with the student's pre-distillation closed-book preference for the original answer. A factorized study of teacher targets, privileged information, divergence, and supervision density finds that explicitly supplying the teacher with annotated answer-critical spans yields the largest joint improvement in evidence response and claim support. These findings motivate Evidence-Paired OPD (EP-OPD), which combines on-policy supervision on matched documents supporting different answers to the same question with an answer-level margin that favors each document's supported answer over its paired alternative. Across both datasets and two student sizes, EP-OPD recovers 54-66% of standard OPD's counterfactual-accuracy loss while preserving 80-85% of its original-document accuracy gain. Our results establish evidence responsiveness as a distinct training and evaluation target alongside answer accuracy. Our code is publicly available at: https://github.com/RuoranXu/EP-OPD.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.01954

Towards Self-Evolving Narrative Systems: Real-Time Adaptation and Long-Term Optimization with LLMs

Ruisheng Sun, Jiaqi Shao, Weijia Han, Linxi Ju, Yunzhen Zhang, Binyan Sun, Rong Chen, Benjamin Blas Maristany 等 12 位作者

原文摘要

Traditional interactive narrative systems offer strong authorial control and structural coherence but often restrict player agency. In contrast, LLM-driven narratives enable open-ended generation and greater flexibility, yet frequently suffer from inconsistency and diminished narrative control. To navigate this trade-off, we propose a hybrid narrative framework that dynamically balances pre-authored content with real-time LLM generation via semantic similarity alignment. Additionally, sustaining narrative quality over extended gameplay remains a core challenge in LLM-based storytelling. To address this, we introduce a self-evolving, feedback-driven optimization framework that leverages in-context learning to iteratively refine prompts, eliminating the need for manual prompt engineering. Through a case study and participant-based evaluation, we show that our framework enhances narrative coherence while preserving player agency, effectively aligning generated content with both authorial intent and evolving player behavior.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.01945

Tracing How Language Models Integrate Temporal Anchors into Factual Recall

Ayana Niwa, Masahiro Kaneko, Kentaro Inui

原文摘要

Factual recall in language models is typically framed as retrieving a factual association: a subject--relation pair maps to a single object. However, many facts are context-dependent: the same pair maps to different objects depending on a condition such as time. We introduce anchored factual recall, in which the model receives an explicit conditioning expression, an anchor, and must select the object valid under it. This raises questions that single-object recall does not: where the anchor enters the computation, what it changes, and how that change selects the answer. Focusing on temporal anchors, we answer them with causal interventions on four open-weight models on English prompts and identify two routes. In a subject-mediated route, the anchor resolves the final subject-token representation to the queried year, and the relation redistributes the readout of this state across attention heads. In a direct route] the final position attends to the anchor, and upper-layer MLPs turn this signal into a preference for the time-appropriate object. The two routes are parallel channels: their contributions add up, most queries use both, and their balance depends on the query, e.g., on the relation, consistently across models. This reframes mechanistic accounts of factual recall from retrieving a single stored object to selecting among context-dependent alternatives.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.10092

RAP: Research Attention Prediction in Language Agents

Yingqian Wu, Jingcong Liang, Shengbin Yue, Siyuan Wang, Zhenfei Yin, Junchi Yu, Zhongyu Wei

原文摘要

LLM agents increasingly support research planning and execution, where judgments about current and anticipated research activity inform decisions. Yet whether they can infer changes across a whole field from the partial evidence they retrieve through online scholarly search remains unclear. We introduce Research Attention Prediction (RAP), a literature search benchmark for evaluating this ability. We build a local search environment from 356,357 dated arXiv papers and construct two complementary evaluation tasks. The benchmark comprises a Core set of 2,160 questions across 54 AI/ML fields and an Extended set covering 269 fields. Our evaluation and analysis of LLM agents show that search substantially improves stronger agents' recognition of recent changes, while forecasting gains remain limited or negative. Agents update forecasts across dates and adjust their investigations to forecasting goals, but these adaptations do not reliably yield accurate predictions of future shifts. Even for recent changes, final judgments can diverge from what the retrieved papers indicate. These findings show that agents can recognise some recent changes, yet reliably anticipating future research activity remains difficult despite changes in their answers and investigations.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗