跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+979新增论文
+781其中 AI 相关
+119已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-20 08:48 · 全站最近入库:2026-11-20 19:58各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 09:14 — 2026-11-21 09:14(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:2026-11-20 08:48最近发现新增版本:2026-11-20 19:58库内论文最新更新时间:2026-11-20 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。

28,961 已收录论文24,030 AI 相关8,912 有修订112 HF 社区发现

浏览论文库

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 144 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 08:46(北京时间)
版本更新 · v2arXiv:2605.30727

MosaicLeaks: Privacy Risks in Querying-in-the-Open for Deep Research Agents

Alexander Gurung, Spandana Gella, Alexandre Drouin, Issam H. Laradji, Perouz Taslakian, Rafael Pardinas

原文摘要

Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent's external queries may leak sensitive information from its local context. This risk is amplified by the mosaic effect, where individual queries may appear harmless but become revealing in aggregate. We introduce MosaicLeaks, a benchmark of 1,001 multi-hop deep research tasks that chain private enterprise documents and a public web corpus, forcing agents to make external queries that depend on local information. We evaluate leakage with an adversary LLM that observes only the agent's external queries and attempts to infer private information at three levels: the agent's research intent, answers to specific private questions and verifiable claims about the enterprise documents. We find that models across families and sizes frequently leak at all three levels, that zero-shot privacy prompting has inconsistent effects across models and that reinforcement learning for task performance alone worsens leakage. To address this, we propose Privacy-Aware Deep Research (PA-DR), an RL framework that combines situational rewards for task success with a learned privacy classifier to provide dense credit assignment over both per-query and mosaic-level leakage. Training Qwen3-4B-Instruct with PA-DR improves accuracy from 48.7% to 58.1% and reduces Privacy Leakage from 34.0% to 6.5%.

cs.CL

版本更新 · v2arXiv: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

新论文 · v1arXiv:2611.25433

RoCoKV: Accelerating LLM Inference Serving through Uncertainty and Redundancy Aware KV Cache Compression

Payman Behnam, Elaheh Sobhani, Saad Amin, Yahya Hassan, Daniyal Khan, Biswadeep Chakraborty, Tanish Patwa, Souvik Kundu 等 9 位作者

原文摘要

KV cache compression is critical for large reasoning models, where most of the cost comes from long generated reasoning traces rather than the input prompt. These traces often contain redundant and low-utility tokens due to looping, self-verification, and corrections. However, such tokens may still receive high attention scores, making it difficult to remove them without harming the reasoning process. We propose RoCoKV, a KV cache compression method designed for large reasoning models. RoCoKV introduces two key components. First, uncertainty-guided adaptive pruning adjusts the compression strength based on the model's confidence, allocating more memory during uncertain phases and applying stronger pruning when the model is confident. Second, redundancy-aware token selection reduces the impact of tokens that are highly similar to recent reasoning content, encouraging the model to retain more informative and diverse tokens. Beyond the algorithm, we co-design RoCoKV with the serving stack: we develop fused three-phase CUDA decode kernels and integrate them into vLLM as a drop-in plugin that preserves its paged KV layout and CUDA graph capture, turning algorithmic savings into real end-to-end speedups. Extensive experiments across three reasoning benchmarks and two reasoning models demonstrate that RoCoKV consistently improves reasoning accuracy under constrained KV cache budgets, achieving higher accuracy compared to state-of-the-art KV cache compression methods. For DeepSeek-R1-Distill-Llama-8B in vLLM, RoCoKV achieves a 1.27X speedup and 38% higher throughput on AIME25, the most challenging evaluated reasoning benchmark, and a 1.43X speedup and 45% higher throughput on the long synthetic benchmark, compared with Full-KV.

cs.CL

新论文 · v1arXiv:2611.24711

Openhagen Worlds: Calibrated Environments for Measuring and Training Scientific Discovery Agents

Alexander Boesgaard Lorup

原文摘要

Agents that do science must choose informative experiments and reason from their results, but most discovery benchmarks cannot separate that skill from recall, unstated scoring rules or shortcut policies. We introduce Openhagen Worlds, a suite of 27 procedurally generated environments (some with finite truth spaces), from classic puzzles of experimental reasoning to simulator-grounded physics and chemistry, in which an agent runs experiments under a budget and commits an answer that a deterministic verifier checks against a hidden truth. Every observation states the complete scoring rule, and each of the 178 difficulty levels is calibrated: it is admitted only if, on 64 development instances, a scripted oracle succeeds on at least 0.75, the strongest plausible non-deductive policy on at most 0.25 and each gated shortcut on at most 0.35. The 92 core levels span all 27 families and are sized to train small models; the 86 frontier rungs extend the 15 discovery worlds with harder questions and with campaigns of three, five and eight or more committed stages on one hidden system. On the same 84 withheld frontier instances, one per family and rung, DeepSeek V4.1 Flash solves 32 at its default reasoning effort with 16,384 tokens per call and 57 at its maximum effort with up to 393,216 tokens per call (oracle: 79); the headroom that remains is largest in eight-stage campaigns. On-policy expert iteration gives a 9B model modest gains: its core-tier success on trained families rises from 0.090 to 0.148, partly through better action format, with no detectable loss on GPQA Diamond. A hosted service plays the core tier's withheld instances.

cs.LG · cs.AI · cs.CL

版本更新 · v2arXiv:2510.05364

Challenging Quadratic Attention$\unicode{x2013}$A Holistic View on the Rise of Alternative Language Model Architectures

Alexander M. Fichtl, Jeremias Bohn, Josefin Kelber, Edoardo Mosca, Georg Groh

原文摘要

Transformers have dominated sequence processing tasks for the past seven years$\unicode{x2013}$most notably language modeling. However, the inherent quadratic complexity of their attention mechanism remains a significant bottleneck as context length increases. We review and distill the recent efforts to overcome this bottleneck, including advances in (sub-quadratic) attention variants, recurrent neural networks, state space models, and hybrid architectures. We critically analyze approaches regarding compute and memory complexity, benchmark results, and fundamental limitations to assess whether the dominance of pure-attention transformers may soon be challenged, which we consider possible, particularly in domain-specific and edge-device applications.

cs.CL

版本更新 · v3arXiv:2602.16169

Unlocking Continuous Diffusion Language Models via Discrete Stochastic Localization

Yunshu Wu, Jiayi Cheng, Longxuan Yu, Partha Thakuria, Rob Brekelmans, Evangelos E. Papalexakis, Greg Ver Steeg

原文摘要

Continuous diffusion is a natural framework for non-autoregressive generation but has generally lagged behind masked discrete diffusion models (MDMs) on discrete sequence generation. We argue that continuous diffusion can effectively model discrete sequences through the joint design of noise parameterization and token representation geometry. We introduce \emph{Discrete Stochastic Localization} (DSL), a continuous-state framework that combines a stochastic localization channel with unit-sphere token embeddings, resulting in a Bayes-optimal denoiser that depends only on the noisy state, without additional timestep conditioning. A single trained denoising network then supports multiple sampling methods, from continuous denoising to masked diffusion, random-order autoregressive generation and hybrid continuous-then-discrete generation. Parallel generation in this continuous-state framework, therefore, need not imply synchronized denoising: some tokens can establish context earlier while others remain open to refinement. Fine-tuning a pretrained MDLM checkpoint with DSL substantially improves generation quality, as measured by MAUVE, on OpenWebText across all step budgets from $T{=}128$ to $T{=}1024$. The same checkpoint supports random-order autoregressive sampling, and a hybrid continuous-then-discrete sampler using as few as $T=48$ total steps, without distillation or further training.

cs.LG · cs.CL

新论文 · v1arXiv:2611.23704

Localizing Text Anonymization for Trustworthy AI: Extending RAT-Bench to Malaysian Microdata and PII

David Hong Liang Chew, Zexi Yao, Nataša Krčo, Matthieu Meeus, Waqas Khalid Obeidy, Yves-Alexandre de Montjoye

原文摘要

As governments and organizations adopt foundation models for public services and workplace productivity, sensitive citizen, employee, and administrative text may enter LLM workflows through inference, retrieval, fine-tuning, or local model development. We examine whether text anonymizers transfer to new deployment settings by extending RAT-Bench to Malaysia using local microdata, Malaysian PII formats, and culturally grounded transcripts in Malaysian English and Bahasa Malaysia. We evaluate NER- and LLM-based anonymizers using an LLM attacker that infers attributes from anonymized text, measuring both re-identification success and text utility. Across Malaysian English and Bahasa Malaysia, the two LLM anonymizers provide the strongest explicit privacy-utility trade-offs, reducing Easy/Hard re-identification risk to 23-29% while preserving BLEU scores of 0.77-0.94. Non-LLM tools show sharper failure modes, either preserving utility while leaving high residual risk or reducing risk through severe over-redaction, especially in Bahasa Malaysia. These findings suggest that anonymizers should be evaluated against deployment-relevant languages, identifier formats, and cultural contexts before being relied on in public-sector, workplace, or local AI pipelines.

cs.CL

版本更新 · v3arXiv:2606.19336

Learning User Simulators with Turing Rewards

Yingshan Susan Wang, Cedegao E. Zhang, Linlu Qiu, Zexue He, Pengyuan Li, Alex Pentland, Roger P. Levy, Yoon Kim

原文摘要

Learning to simulate human users in interactive settings could advance the training of agent assistants, evaluation of personalization systems, research in the social sciences, and more. Existing approaches generally do so by training a large language model (LLM) to match a single ground truth response, either by maximizing the log probability or by using a similarity reward. We instead propose Turing-RL: a Turing-Test-based reinforcement learning approach for training user simulator models. Turing-RL uses a discriminative Turing reward with an LLM judge to score how indistinguishable a generated response is from the real user's given the user's history, and the user simulator LLM learns to produce responses indistinguishable from what the user could have said with such rewards. Across two different domains--conversational chat and Reddit forum discussion--we find that Turing-RL consistently outperforms baseline methods on both LLM and human evaluation metrics. Our study suggests that optimizing for indistinguishability, rather than response matching, is effective for learning user simulators.

cs.CL

新论文 · v1arXiv:2611.23359

Prioritize, Don't Decide: What We Learned Auditing a Deployed Human-in-the-Loop CV Sorter

Hadi Mohammadi, Jeroen Sparla, Alexandros Paliouras, Georg Krempl, Robert A. Bagheri

原文摘要

A single academic vacancy can draw hundreds of applications, and the first screening round consumes most of a committee's time. We describe a CV Priority Sorter deployed at a national platform for academic vacancies and integrated with its vacancy database. It proposes a reading order for a pile of CVs and leaves every decision to the committee, pairing a frozen multilingual transformer encoder with a per-session linear model that retrains after each label and is discarded when the session ends. It works: on a pool of 300 synthetic CVs, a committee reading in the tool's order reaches 95% of the relevant candidates after half the pile, where an arbitrary order needs almost all of it. This paper is about what we found when we audited that claim, rather than about the system itself. Three findings changed the product. The pool turned out to be sortable by document length alone, which makes precision-style metrics on it uninformative; regenerating it under a shared length budget removes that shortcut and leaves the reading-depth result standing. The four-level agreement score such systems report, ours included, is a binary probability cut at fixed thresholds, so it follows the classifier's confidence rather than its ranking and falls as labels accumulate. And uncertainty sampling, the obvious way to implement a "suggest the next CV" button, leaves model quality intact but raises the committee's total reading load above the no-label baseline, because the CVs it asks about are the ones that would otherwise be skipped. A fourth result shows where to look next: a hosted LLM embedding model lifts zero-label ordinal discrimination from 0.66 to 0.80 under length control, then stops mattering once the session model has two dozen labels; a local cross-encoder matches that cold-start gain without sending data out. We report the fixes each finding forced.

cs.CL · cs.IR · cs.LG

版本更新 · v2arXiv:2609.34839

OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations

Faadil Mustun, Chiara Semenzin, Roberto Dessi, Pablo Robin Guerrero, Pierre Orhan, Alexis Emanuelli, Emanuele Rossi, Yair Lakretz 等 10 位作者

原文摘要

Recent advances in bioacoustics have been driven by large-scale corpora and standardized benchmarks, yet existing resources are overwhelmingly bird-centric and shallow per species, limiting their use for studying the structure of a single species' communication system. This gap is particularly acute for cetaceans: despite bottlenose dolphins (Tursiops truncatus) being a compelling case of complex vocal communication among non-human mammals, existing dolphin datasets are small, fragmented, and largely closed. We introduce OpenWhistle, the largest publicly available dataset of dolphin vocalizations. It comprises approximately 180,000 whistles (114 hours) recorded over five years from a stable pod of five individuals in a semi-natural environment, paired with a curated subset of 8,354 expert-annotated whistles and reproducible evaluation protocols for whistle-type detection and classification. We further release the full processing pipeline for whistle detection, segmentation, and categorization. To demonstrate its utility, we pretrain a Wav2Vec2.0 model adapted to dolphin acoustics on the OpenWhistle corpus and show that it learns effective representations, outperforming general-purpose bioacoustic models such as AVES and BioLingual on both tasks while leaving meaningful headroom for future work. By releasing the dataset, pipeline, and evaluation protocol, we provide the first open dolphin whistle dataset tailored for training self-supervised models, laying the groundwork for advancing dolphin communication research and developing models that capture fine-grained acoustic structure within species.

cs.CL

新论文 · v1arXiv:2611.21449

Replicating and Extending Weight-Space Programming: Frontier LLMs Compile Programs into Weights and Read Input Layers

Roksana Goworek, Alexander Hickey, Federico Nanni, Martin O'Reilly

原文摘要

Researchers at MIT recently reported that GPT-6 Astra, a frontier large language model (LLM), can write the weights of small neural networks directly in a single reply, and can read the raw weights of small networks to say what they were trained to do. In a three-day sprint, we replicated and extended these results. Our four findings: (1) Writing weights reproduces. Astra's networks for the game Breakout scored 514, 550 and 307 (versus MIT's 517), and scored just as well on two modified versions of Breakout. Claude Opus 5.5 can also write working networks, but often runs out of output tokens. (2) What is new is translating programs into weights. Earlier models could write a good policy as code a generation before they could write network weights, and the networks the models write behave like compiled programs: two models independently decompiled Astra's Breakout network into Python matching its actions in all tested states. On classifying handwritten digits, Astra's networks scored 50% on average, below a simple classifier (82%). (3) Reading weights is mostly reading the input layer. Astra and Opus 5.5 identify which game a network plays almost perfectly, but on three new games we designed and trained ourselves, Astra recovered where things are on the board, while mostly missing how the game works. (4) Auditing from weights alone is possible but shallow. Without executing code, Astra picked out a small language model poisoned during training in 10 of 10 runs, one poisoned by extra training afterwards in 7 of 10, and a malicious adapter in 23 of 32 cases, but never identified the trigger or malicious behaviour. Open-weight models lag behind. From this early study, we conclude that LLMs can read and write small networks, but that this currently amounts to compiling hand-designed programs and reading the layout of the input layer. It is a promising but early tool for interpretability.

cs.CL · cs.AI

新论文 · v1arXiv:2611.21422

Rubric-Based AES with LLM Ensembles: Scoring of Student Responses in Critical Online Reasoning Using Ensembles of LLMs

Daniel Baumartz, Alexander Mehler, Patrick Schrottenbacher, Leon Hammerla, Sebastian Gombert

原文摘要

Grading student responses is costly and time-consuming. Moreover, human annotators are prone to bias, fatigue, and rater drift, which further limits grading reliability. Various methods have been employed to develop automated essay scoring (AES) systems to reduce these costs and address bias-related issues. These methods range from early text-feature-based systems to current machine-learning and transformer-based models. Specifically, large language models (LLMs) offer the potential to streamline the development of AES systems. However, they are subject to well-known limitations, such as hallucinations and prompt sensitivity. This creates a need to explore the potential of a mixed-methods approach that leverages the synergies among individual LLMs through ensemble formation. We evaluate the potential of ensembles of LLMs to match human-rater agreement when grading student responses in economics and medical education across 103 criteria. We use 631 German-language responses and corresponding human ratings by 2 to 5 annotators from a previously conducted assessment. We experiment with 523 rubric-based prompt configurations across 15 open-weight LLMs and assess their consistency with human raters using Krippendorff's alpha for inter-annotator agreement (IAA) and the Alternative Annotator Test (alt-test). Our results show that replacing one human rater with the best-performing LLM--prompt configuration maintains or improves IAA for 91 of the 103 criteria and that LLMs pass the alt-test for 90 criteria. Our findings suggest that current state-of-the-art LLMs can support AES of German responses in economics and medical education, reducing human workload and enabling faster processing, while validation against human ratings remains necessary.

cs.CL

新论文 · v1arXiv:2611.21376

A Crowd-Audited Atlas of Implicit Bias in LLM-Powered Agents

Jason Alexander, Yuxuan Li, Michael Licamele, Wesley Hanwen Deng, Sauvik Das

原文摘要

Prior work has shown that large language models (LLMs) express implicit biases in high-publicity, one-off decision scenarios such as hiring and intelligence assessment. However, as people increasingly employ autonomous, LLM-powered AI agents that make decisions on their behalf in doing day-to-day tasks, it is of critical importance that we understand how implicit biases, based on sociodemographic assumptions, emerge across a broad milieu of workaday decision scenarios. We used the 17 domains of daily activity in the American Time Use Survey (ATUS) to generate an atlas of 506 decision scenarios in which LLMs --- endowed with personas representing different races, gender identities, and political ideologies --- expressed significant implicit bias. Next, in a crowdsourced audit, 500 participants made 8,010 binary choices about which implicitly biased decision scenarios most deserved further review. We found that participants prioritized decision scenarios that had broad relevance (e.g., about food choices), were high stakes (e.g., about health), and where all members of a simulated sociodemographic group made unanimous decisions (because it was a clear indication of stereotyping). They were less swayed by the magnitude of an observed disparity or the specific ATUS domain the decision scenario represented. When examining the rationale behind their prioritization choices, we found that participants often interpreted disparities as traits of demographic groups and judged whether those traits seemed plausible, often based on their own internalized stereotypes. We discuss implications for building personalized AI agents, digital twins, social simulations, and for using crowdsourced bias audits.

cs.CY · cs.CL · cs.HC

新论文 · v1arXiv:2611.21208

Validating LLM-Based Essay Scorers: A Comparative Analysis in Critical Online Reasoning in Medical Education

Daniel Baumartz, Alexander Mehler, Mevlüt Bagci, Maxim Konca

原文摘要

Automated essay scorers (AES) based on large language models (LLMs) promise human-level accuracy. However, training, fine-tuning, and evaluating LLMs for specific assessment tasks is costly. In particular, the question arises whether these AES should be trained for specific domains or whether models based on generic LLMs might be sufficient. Human annotations, by contrast, are often more valid but far more expensive. This study evaluates whether ensembles of AES systems based on general-purpose LLMs can approximate human ratings of medical-education writing tasks by leveraging encoded linguistic and conceptual knowledge. The goal is to assess whether general-purpose models are applicable for assisting or potentially replacing human annotation in this domain. We experiment with 11 AES systems and 8 different LLMs and evaluate 668 student responses to 12 tasks, using \numprint{2076} evaluations by 7 human annotators. These ratings are aggregated to create consistent performance levels that can be used as target classes for classifiers trained on AES outputs. Our results show that learner classes based on human ratings can be accurately reconstructed using a linear SVM trained on a small set of AES features. A single classifier generalized well across tasks and domains, with reasoning and content features being most predictive. This suggests that out-of-the-box LLMs, even without domain-specific fine-tuning, can support automatic essay grading in medical education. While individual AES systems are insufficient alone, their combination approximates human evaluators, pointing to a reduced need for manual scoring or its supplementation with automated methods.

cs.CL

新论文 · v1arXiv:2611.20803

What Do Digital Traces Predict? Task Effects, Student Differences, and Critical Online Reasoning

Maxim Konca, Alexander Mehler

原文摘要

Digital assessment traces are increasingly used to predict student performance, often evaluated by correlating predicted and observed scores pooled across tasks. However, such correlations may reflect between-task differences rather than performance differences among students within tasks. To investigate this, we analyzed up to 527 browsing sessions from 424 university students completing open-Internet Critical Online Reasoning tasks and assessed the cross-task stability of their ratings. We evaluated the same out-of-fold predictions within and across tasks, compared three validation tiers up to leave-one-task-out, and tested incremental predictive contributions of consultation order and relational structure against matched permutation and randomization controls. None of the twelve prespecified primary tests was significant: on held-out tasks, consulted-page features showed no significant association with relative student performance, nor did observed order or relational structure add predictive value. Pooling observations across tasks substantially increased correlations, particularly for factual correctness, whose scores showed limited evidence of stable between-student differences across tasks. However, sentence embeddings yielded modest positive correlations with generic response quality on held-out tasks, without establishing added predictive value from observed order or relational structure. Our findings show that pooled predictive correlations may not reflect within-task student differences, highlighting the need for within-task evaluation and criterion generalizability audits.

cs.CL · cs.CY · cs.LG

新论文 · v1arXiv:2611.20771

Nemotron 3.5 Bolt E4B 4GB: An Open and Efficient Small Language Model for Fast Conversational Agents

NVIDIA, :, Adi Margolin, Aditya Malte, Aishwarya Chhabra, Alex Hodson, Ameya Sunil Mahabaleshwarkar, Ambrish Dantrey 等 113 位作者

原文摘要

We introduce Nemotron 3.5 Bolt E4B 4GB, an open and efficient small language model further post-trained from Gemma-4-E4B-it and optimized for speed and intelligence per GB of memory used for fast conversational agents. The model carries forward Gemma's strong conversational capabilities and further improves multi-turn tool use, instruction following, and input grounding for fast, local voice/device assistants, interactive personas, and embedded/Edge interfaces. The model combines natural conversation with multi-turn agentic capabilities to power voice/device assistants, interactive characters and robotic/IoT interfaces running on GeForce RTX, Jetson, and RTX/DGX Spark. Experiments on 21 benchmarks show that Nemotron 3.5 Bolt E4B 4GB achieves the highest average accuracy while generating the fewest output tokens in both reasoning and non-reasoning modes, compared with Gemma-4-E4B-it and Qwen3.5-4B. The model is released with a DFlash drafter. With DFlash, Bolt E4B achieves $2.74\times$ the per-user decode speed of inference without speculative decoding at concurrency~1, and $1.21\times$ the generation throughput of Gemma-4-E4B-it with MTP at batch size 1.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.20156

Risk-certified prompt optimization for LLM-as-a-judge guardrails

Hussein Jawad, Alexandre Landeau, Guy Fleury Iradukunda, Paul Beaumont, Théodore Rezel, Sofiane Schaack, Nicolas J-B. Brunel

原文摘要

LLM-as-a-judge guardrails are often controlled primarily through system prompts, yet standard prompt optimization methods target average performance rather than explicit safety constraints. This paper studies system-prompt selection as a constrained optimization problem: maximize benign pass rate while keeping the missed-attack rate below a user-specified level α with finite-sample statistical control. The framework combines LLM-based judging, automatic prompt generation, and distribution-free risk certification. This work instantiates it in two procedures: CPS and CRISP. CPS generates a fixed pool of prompts from training data, then certifies candidates on a held-out evaluation set using simultaneous upper confidence bounds and selects the certified prompt with the highest benign utility. CRISP instead integrates risk feedback into an adaptive search, using hard example-guided rewriting and a reusable-holdout interface to steer generation toward the feasible utility frontier. Experiments on four datasets covering prompt injection, system-prompt leakage, and sensitive-information disclosure show that risk-certified selection consistently respects the target missed-attack level on heldout tests, while utility improves as the permitted risk is relaxed. CRISP is especially useful when fixed-pool search fails to produce certifiable high-utility candidates.

cs.CL · cs.LG

新论文 · v1arXiv:2611.20094

FRAAC: A French Multimodal Large Language Model and Dataset forPictogram-Based Augmentative and Alternative Communication

Alexandre Audibert, François Portet, Benoit Favre, Maja Jonck Hjuler

原文摘要

Pictograms are a core modality in augmentative and alternative communication (AAC), yet recent multimodal large language models (MLLMs) cannot reliably interpret or produce pictogram sequences. Two obstacles explain this gap: the absence of a standardized pictogram grammar and the scarcity of parallel text-pictogram data. We introduce Retrieval-Augmented Pictogram Selection (RAPS), an annotation method that refines a rule-based transcription system's output with LLM-based concept extraction and retrieval over a pictogram vocabulary, handling cases where hand-written rules fail (named entities, numbers, polysemous words) without requiring new rules. Applied to French, RAPS yields a synthetic corpus of 1.4 million parallel pairs, of which 300k are publicly released. In a human ranking study where annotators order four pictogram transcriptions of the same sentence, RAPS outperforms the rule-based state of the art. Moreover, we show that standard reference-based metrics separate systems of clearly different quality but fail to rank strong systems against each other. Finally, we build an instruction-tuned model that supports interaction in several AAC modalities: it accepts pictogram or text input and answers in pictograms, speech, or text.

cs.CL · cs.AI

版本更新 · v2arXiv:2611.07940

CardioLM: A multilingual suite of small language models for the cardiology domain

Manuela-Daniela Danu, Miguel Rodríguez-Ortega, Alberto Becerra-Tome, Pablo Arancibia, Fernando Gallego-Donoso, Lorenzo Pratesi, Xi Bai, Andrei-Damian Dinescu 等 40 位作者

原文摘要

Background: Cardiovascular care generates large volumes of clinical free-text documentation in local languages. However, clinical natural language processing (NLP) resources remain concentrated in English, and no existing benchmark is cardiology-specific, multilingual, and broad in entity coverage. Methods: We present two contributions. The first is CardioCCC, a corpus of 508 cardiology case reports originally published in Spanish and extended to Czech, Dutch, English, Italian, Romanian, and Swedish through machine translation, annotation projection, and clinician review, with annotations covering diseases, symptoms, procedures, and medications. The second is CardioLM, comprising CardioBERTa, a set of language-specific encoders obtained by continued pretraining on cardiology and biomedical corpora, and CardioNER, a set of named entity recognition (NER) models fine-tuned on CardioCCC. Results: Cardiology-specific pretraining reduced pseudo-perplexity by up to a factor of eight, yet changed strict span-level F1 by at most 0.04, with benefits concentrated in languages whose backbones lacked biomedical coverage. The jointly trained multilingual XLM-RoBERTa-large models achieved the best strict F1 in nearly every language and entity type, exceeding the strongest monolingual system by up to 0.13 F1. A large share of residual errors concerned boundary detection: relaxed matching recovered 0.10-0.23 F1 for disease, procedure, and symptom, but only 0.01-0.07 for medication. Clinicians rated CardioNER well above a rule- and dictionary-based baseline, with the mean rating across entity types rising from 26% to 73%. Conclusions: Specialty-specific pretraining is not the decisive factor for downstream entity recognition, whereas multilingual joint training yields the strongest results in nearly all settings and a simpler deployment path for multi-country clinical information extraction.

cs.CL · cs.AI

版本更新 · v2arXiv:2607.04688

URSA:面向实用性逆合成评估的化学感知基准

URSA: Chemistry-Aware Benchmark for Utilitarian Retrosynthesis Assessment

Bogdan Zagribelnyy, Ivan Ilin, Nikita Bondarev, Anton Morgunov, Arkadii Lin, Maksim Kuznetsov, Rim Shayakhmetov, Vladimir Aladinskiy 等 10 位作者

AI 速读 基于摘要

作者提出URSA评估框架,从化学合理性角度对专用逆合成系统与通用大语言模型的合成路线规划能力进行基准比较。

cs.LG · cs.AI · cs.CE · cs.CL

查看原文摘要

Synthesis planning aiming to find pathways of reactions for a target molecule is one of the most important and challenging tasks in drug discovery. Recent progress has produced both specialized deep-learning retrosynthesis systems and general-purpose large language models, but objective comparison remains difficult due to the lack of flexible, chemically interpretable benchmarking protocols. In the current study, we are introducing the URSA (Utilitarian RetroSynthesis Assessment) evaluation framework that provides the opportunity to benchmark the synthetic routes not only from a formal perspective, such as convergence to commercially available starting materials, but also from a chemical plausibility perspective, mimicking the way expert chemists evaluate the reactions and routes. The study covers a comprehensive evaluation of both conventional end-to-end retrosynthesis solutions and LLMs for the synthesis planning task on a set of novel, diverse target molecules with undisclosed synthetic routes, which represent realistic tasks in the daily drug design routine. Although LLMs can support high-level strategic planning and even achieve SOTA performance in the OOD set, the Pareto frontier for estimated run cost and chemically plausible route solvability consists entirely of specialized retrosynthesis planners.

版本更新 · v2arXiv:2610.21025

Learning to Read Out: Unembedding Dynamics in Language Model Pretraining

Matteo He, William F. Shen, Alex Iacob, Andrej Jovanovic, Xinchi Qiu, Nicholas D. Lane

原文摘要

When does a language model acquire a capability? Developmental studies usually answer this by tracking when models start producing the correct token. But that prediction passes through the unembedding matrix, or readout, which is itself still being learned during pretraining. We show that a model's hidden states can encode task information before its own readout expresses it, so timelines read from outputs conflate what the model encodes with what its readout expresses. To track the readout directly, we introduce parameter-trajectory crosscoding, which fits one sparse dictionary to every token's readout row at every checkpoint. Unlike crosscoders on activations, it needs only checkpoint weights. We use it to locate a narrow early window, 0.4-1.1% of the way through Pythia's pretraining, in which the readout reorganizes and WordNet word categories become decodable from its rows. By then, linear probes on the hidden states already decode task information that the model's own outputs do not yet show. To test whether such gaps depend on how fast the readout learns, we pretrain small models from scratch that differ only in the readout's learning rate. We find that raising it makes the readout develop earlier and the gap close sooner, with little effect on final loss. Together, these results imply that early in training, where many developmental transitions are dated, a behavioral date can reflect when the readout learns to express task information and not when the model learns it.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.18157

A Multi-Layered AI Pipeline for BPMN Generation from Natural Language

Alexandru Rancea, Ionut Anghel, Tudor Cioara, Ovidiu Anchidin, Ancuta Nemes

原文摘要

The automated transformation of unstructured natural language process descriptions into Business Process Model and Notation (BPMN) diagrams remains a challenging task, due to linguistic ambiguity and the difficulty to detect non-adjacent control-flow relationships. This paper proposes a solution for generating BPMN models from natural language descriptions through a multi-layered pipeline that integrates transformer-based architectures, Large Language Models and ontology population. The proposed method starts with natural language descriptions and separates textual processing, control-flow relation extraction, and structural validation into dedicated components. Entity extraction is performed using a fine-tuned RoBERTa-based transformer model while a control-flow engine captures non-adjacent relationships through LLM-based inference. To improve generalization under limited available data, the approach uses a dynamic few-shot strategy with con- trolled template input construction, enabling improved generalization under limited data availability. Additionally, an ontology population mechanism is introduced to represent the generated process models as semantic knowledge structures. For evaluation purposes we have used from the literature and industry reference processes. Additionally, we have validated the proposed method on a real-world heart disease patients transitional care process, capturing practical constraints and multi-actor collaboration, and serving as an expert-defined benchmark scenario. Experimental results demonstrate strong performance in the relation extraction task, achieving an accuracy of 90.7%, as well as maintaining structural alignment between the generated and reference BPMN models across multiple evaluation scenarios scenarios.

cs.CL

版本更新 · v5arXiv:2406.15877

BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions

Terry Yue Zhuo, Minh Chien Vu, Jenny Chim, Han Hu, Wenhao Yu, Ratnadira Widyasari, Imam Nur Bani Yusuf, Haolan Zhan 等 33 位作者

原文摘要

Task automation has been greatly empowered by the recent advances in Large Language Models (LLMs) via Python code, where the tasks ranging from software engineering development to general-purpose reasoning. While current benchmarks have shown that LLMs can solve tasks using programs like human developers, the majority of their evaluations are limited to short and self-contained algorithmic tasks or standalone function calls. Solving challenging and practical tasks requires the capability of utilizing diverse function calls as tools to efficiently implement functionalities like data analysis and web development. In addition, using multiple tools to solve a task needs compositional reasoning by accurately understanding complex instructions. Fulfilling both of these characteristics can pose a great challenge for LLMs.To assess how well LLMs can solve challenging and practical tasks via programs, we introduce BigCodeBench, a benchmark that challenges LLMs to invoke multiple function calls as tools from 139 libraries and 7 domains for 1,140 fine-grained tasks. To evaluate LLMs rigorously, each task encompasses 5.6 test cases with an average branch coverage of 99%. In addition, we propose a natural-language-oriented variant of BigCodeBench, BigCodeBench-Instruct, that automatically transforms the original docstrings into short instructions only with essential information. Our extensive evaluation of 60 LLMs shows that LLMs are not yet capable of following complex instructions to use function calls precisely, with scores up to 60%, significantly lower than the human performance of 97%. The results underscore the need for further advancements in this area.

cs.SE · cs.AI · cs.CL

新论文 · v1arXiv:2611.17955

WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents

Imene Kerboua, Fatemeh Pesaran Zadeh, Xing Han Lù, Weijian Qi, Alexander Miller, Junyi Song, Yunjia Tian, Dongjin Kang 等 22 位作者

原文摘要

Web agents powered by large language and vision-language models are increasingly applied to realistic browser work that spans heterogeneous applications, multimodal content, and stateful workflows. However, existing reproducible web-agent benchmarks cover only a small number of web applications drawn from a few software categories, and restrict modality to text and vision. Live benchmarks broaden site coverage but sacrifice reproducibility, since pages and data drift between runs. Moreover, existing benchmarks do not meaningfully evaluate whether agents can understand and use audio and video content embedded within web tasks. To address these gaps, we introduce WebArena-Pro, a benchmark comprising 300 tasks across 20 self-hosted web applications in six domain categories, spanning distinct interface conventions, workflows, and data models. Across the evaluated agents, the best performance is achieved by both Codex with GPT 6.1 Sol and Claude Code with Claude Opus 5.5 at 62.3%, followed by OpenCode with Kimi 3.1 Pro at 60.7%. The strongest GenericAgent configurations, Gemini 3.1 Pro omni and Gemma4 31B, both achieve 30.3%. Among reproducible, human-curated web agent benchmarks, WebArena-Pro provides the broadest application coverage and the most comprehensive multimodal support to date. The benchmark treats audio and video as core observations alongside text and vision, with dedicated actions for extracting information from each. WebArena-Pro runs each task in isolation and supports reproducible, parallel evaluation. Tasks are authored through a dedicated annotator interface, filtered by LLM-assisted triage, and finally validated by humans before release. Project page: https://webarena-pro.github.io

cs.AI · cs.CL · cs.LG

新论文 · v1arXiv:2611.17931

MineralExp: Benchmarking LLM Agents for Experimental Design in Real-World Mineral Exploration

Bing Wang, Mingyang Wang, Yuanhang Mi, Yue Xu, Kaibo Su, Li'ang Xiao, Chunguo Wu, Changchun Li 等 11 位作者

原文摘要

In mineral exploration, verifying geological hypotheses relies on designing a variety of geochemical experiments. While large language model (LLM) agents increasingly automate downstream geoscientific data analysis, a crucial upstream capability remains largely unexplored: automated geochemical experimental design, i.e., deciding which experiments to conduct before committing to costly laboratory work. Evaluating this capability of LLM agents is difficult for two reasons. First, geochemical experiments exhibit high methodological diversity and subjectivity: to test the same hypothesis, different geologists may choose different minerals and instruments, so no single protocol can serve as the reference answer. Second, open-ended experimental plans resist representation as structured labels, which makes objective scoring difficult. To address these challenges, we introduce MineralExp, a benchmark that reformulates geochemical experimental design as a hypothesis-based experiment selection problem, i.e., a multi-label classification problem. We abstract concrete instrument names and procedural details into a shared taxonomy of 19 experimental types across five mineral-system processes: Source, Transport, Deposition, Modification, and Preservation. This taxonomy yields deterministic multi-label targets while accommodating diverse experimental protocols. Constructed from published ore-deposit literature and verified by domain experts, MineralExp comprises 277 cases and 1,572 reference experimental plans. Given a geological hypothesis and initial context, an LLM agent selects the relevant tasks and generates concrete experimental plans, which are scored by multi-label classification metrics and an LLM-as-a-Judge metric. The experimental results show that current agents generally achieve high recall but low precision, tending to propose more experiments than are necessary to test the hypothesis.

cs.CL · cs.AI · cs.CE

新论文 · v1arXiv:2611.17639

What LLM Agents See and What Users Say

Yue Zhao, Angela Huyue Zhang, S. Alex Yang, Hua Wei, Xiyang Hu

原文摘要

Before an LLM agent cancels a reservation or deletes a folder for a user, what stops it from acting on a false assumption: what its tools show, or what the user says? In this work, we study this gap with presupposition theory from linguistics: an agent may act only if each claim behind the act (e.g., one reservation fits the name) is seen to hold, answered by the user, or disclosed while the act can be undone. A simple bound shows that when lookups cannot tell one record from two, any agent acts without asking first as often with two records as with one. We then vary whether a search lists, counts, or omits a second match, and whether the user states a claim that a record refutes (e.g., "It is empty"). Under a role-only prompt, a count of a second record leaves most of 36 earlier agents acting on the one they see, whereas listing both stops most. When the user says a record such as a folder is empty, however, a count of its one item stops agents about as often as a listing, and stating the claim mostly cuts acting when it is false. The newest Claude and GPT models always open a counted second record, then never act. When we add a same-name, same-zip customer to a retail benchmark, a lookup returning that customer first led to wrong-account writes in about a quarter of conversations, and one that checks a street the user knows kept completion near the one-customer level. Safe writes therefore need lookups that list or check competing records, and evaluations that vary what tools show.

cs.AI · cs.CL

版本更新 · v3arXiv:2608.11426

趋同是否不可避免?将输出同质性追溯至基座模型

Is Convergence Inevitable? Tracing Output Homogeneity Back to Base Models

Alexandrine Fortier, Hazel Chen, Peter West

AI 速读 基于摘要

作者提出语言模型输出同质性可能源于预训练而非对齐,SFT 只是揭示或放大这一现象。

cs.CL

查看原文摘要

The lack of diversity in LM content is widely attributed to the alignment process, but how and where exactly in the pipeline this collapse begins is unknown. We argue that output homogeneity is likely learned during the pretraining phase, and only revealed or magnified during the alignment process. Specifically, we find that semantic convergence is observed from the first alignment stage--the instruction-tuning phase (SFT)--suggesting that homogeneity might already exist in the pre-alignment model. To investigate this, we conduct controlled SFT experiments examining how training data influences output convergence on specific input/output pairs. We find that convergence can be revealed and amplified, but not introduced by the SFT data, supporting its role as a catalyst rather than a cause. To further test whether homogeneity originates before alignment, we measure convergence in base models. We find that instruct-like collapse can be induced through prompting alone, even without alignment. Taken together, our results suggest that semantic convergence may arise naturally from the objectives underlying LM training, making it difficult to mitigate through post-alignment interventions alone.

新论文 · v1arXiv:2611.16929

SPADE:面向AI文本检测的软提示

SPADE: Soft Prompts for Ai-text DEtection

Alex Jiang, Didier Schwab, Chuyuan Li

AI 速读 基于摘要

SPADE 通过度量提示对文本的解释程度(PMI)作为判别信号,为每个类别训练一个软提示,从而提升AI生成文本检测的跨数据集泛化能力。

cs.CL · cs.AI

查看原文摘要

The easy access to Large Language Models (LLMs) has made AI-generated texts on the web increasingly ubiquitous, making AI-text detection a crucial task. Prompts greatly influence the generation process, but existing detectors focus solely on the generated texts --- supervised methods learn from labeled texts only, and zero-shot methods rely on the text's probability distributions --- ignoring the underlying generation process, which limits their robustness in open-world scenarios. We introduce SPADE: Soft Prompts for Ai-text DEtection, a novel framework for detecting AI-generated text that transfers better across datasets. Specifically, we measure the prompt's influence and use it as a discriminative signal. We characterize the prompt's influence by introducing a Pointwise Mutual Information (PMI) term that captures how much a prompt explains a text. Based on this quantity, SPADE trains one soft prompt per class to amplify the separation, and predicts the class whose prompt best explains the candidate text. Extensive experiments on several benchmarks against various detectors demonstrate the robustness and effectiveness of our method, achieving state-of-the-art (SOTA) performance on out-of-distribution (OOD) datasets.

新论文 · v1arXiv:2611.15779

Honesty Is a Race: Predicting When Verifiable-Reward RL Teaches a Model to Bluff

Nathaniel Bottman, Alexei Oblomkov

原文摘要

Two rewards with the same expected payoff at the start of training can end a reinforcement-learning run in opposite places, with an honest model or one that bluffs. We study rewards that pay $b$ for a verdict that a checker cannot verify ("this program is correct") and $r$ for a certificate that it accepts (an input revealing a bug). To first order, a GRPO step makes the model more honest exactly when $\tilde{p}\,r/b>1+κ$. Here $\tilde{p}$ is the rate at which the checker accepts the model's certificates, weighted toward the items on which the model still gives both verdicts, and $κ$ is the pull of the items on which the unverifiable verdict is right. Both terms move during training. A checker raises $\tilde{p}$, so near the threshold honesty is a race, which a run just below the threshold can win and a run under a lottery (random pay with the same mean) cannot. The term $κ$ starts large and falls as the other items settle, so a run just above the threshold first dips toward bluffing, on code nearly to zero, before climbing back. Paying the honest verdict also pulls the other items toward it, which predicts a ceiling on the payoff ratio, above which the model gives the honest verdict where it is wrong. On propositional Datalog, two code models, and abstention on grade-school math, with $p$ the model's success rate before RL, no run above $pr/b=1$ ends collapsed, and runs under the checker can survive slightly below it. At $pr/b=3$, above the ceiling, both code models call about half of correct programs buggy. On depth-3 Datalog, training generously under the checker and then cutting the payoff keeps the model at least 67% honest in three seeds, at a ratio where the same model trained from the start falls to 4-15% and is at most 47% honest after 160 steps. On depth-2 Datalog, when honesty collapses, a linear probe fit before RL still reads the truth.

cs.LG · cs.AI · cs.CL

版本更新 · v2arXiv:2611.13498

Evaluating Compositional Generalization in Transformers: The Role of Composition Equivalence and Module Coverage

Purva Pruthi, Andrew Yuan, Alexander D'Amour, David Jensen

原文摘要

Compositional generalization is an important form of out-of-distribution generalization, defined as the ability to train on some combinations of modules (e.g., functions such as sort, filter) and then generalize to unseen combinations. Despite substantial empirical work evaluating compositional generalization in transformers, findings have been inconsistent because evaluations were often conducted on a limited set of train-test splits, defined in an ad hoc manner. Through a systematic evaluation, we identify two key aspects of the train-test distribution that strongly affect performance: composition equivalence and module coverage. We show that the apparent performance of direct models (trained only on final outputs) results primarily from exploiting composition equivalences, meaning different sequences of modules that reduce to identical end-to-end functions. When benchmarks eliminate these equivalences, performance drops significantly, showing their inability to generalize to compositions of known modules that produce novel end-to-end functions. We further demonstrate that performance degrades monotonically as module coverage shifts increase (when modules appear in different contexts between train and test), affecting both direct and step-by-step models (trained on intermediate and final outputs). These findings have concrete implications for benchmark design: benchmarks that include compositional equivalences can lead to significant overestimation of the generalization abilities of direct models, and benchmarks that include both module coverage violations and compositional equivalences can exacerbate shortcut learning. As a case study, we demonstrate the role of compositional equivalence on the generalization performance of the pre-trained model Llama3-8B on the code-output prediction task in the coding benchmark LiveCodeBench.

cs.LG · cs.CL