Language_Model
-
Do Language Models Feel the Sentiment? Investigating Emotion Vectors
2026-11-10 08:45:50Ali Hassan Khan, Saad Bin Ather, Muhammad Saif, Maryam Bashir, Hajra WaheedAbstract
Emotion vectors are directions in a language model's internal representations that are associated with emotional concepts. Prior work shows that these vectors can influence model behavior, but their role in sentiment classification remains unclear. This study investigates emotion directions in Qwen2.5-0.5B-Instruct and Qwen2.5-3B-Instruct. Eight directions extracted from GoEmotions are evaluated through sentiment probing, activation interventions, and supervised correction. Emotion-only probes achieve 88.66% and 90.70% accuracy on binary TweetEval. In exploratory matched-ablation experiments, raw emotion-space removal causes 2.39 and 1.10 percentage points more accuracy loss than matched isotropic random-subspace removal for 0.5B and 3B, respectively. All four TweetEval raw and denoised effects are significant after Holm adjustment across eight comparisons. An exploratory SST-2 extension shows smaller raw excess losses of 0.66 and 0.57 points, with no effects surviving Holm adjustment. Frozen sentiment-correction classifiers are also tested on 20,000 Amazon Polarity reviews. Emotion-based correction improves accuracy over uncorrected Qwen, but does not consistently outperform margin-only baselines. These results distinguish sentiment information available in model representations, the effects of changing those representations, and their practical value for classification. They do not establish a uniquely emotional mechanism.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13744
PDF
-
Toward Autonomous RTL and Synthesis Optimization: Benchmarking and Self-Improving Agents
2026-11-10 08:07:31Wenji Fang, Yao Lu, Jinghua Wang, Zhiyao XieAbstract
Recent advances in large language models (LLMs) and autonomous agents are creating new opportunities for hardware design optimization. This paper examines three connected directions toward autonomous RTL and synthesis optimization: (1) Realistic benchmarking. We discuss reference-based and open-ended RTL optimization benchmarks that evaluate optimization opportunities under realistic EDA workflows. (2) Tool-grounded RTL optimization. We present Dr. RTL, a multi-agent framework that automates critical-path analysis, RTL rewriting, and implementation-level evaluation, while continuously improving through reusable optimization skills. Evaluated on 20 real-world RTL designs, Dr.~RTL achieves average WNS/TNS improvements of 21\%/17\% with a 6\% area reduction over commercial synthesis. (3) Design-specific synthesis optimization. We explore agent-driven synthesis script tuning and the discovery of executable synthesis algorithms across intermediate representations, technology-independent optimization, and technology mapping. Finally, we envision AI-native design automation that jointly explores architectural, RTL, and EDA algorithm design spaces, guided by verification constraints and implementation feedback.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13713
PDF
-
On Private Knowledge Retrieval and Private Inference for secure multi-agent systems
2026-11-10 07:43:02Vicen\c{c} Torra, Maria Bras-Amor\'osAbstract
Multi-agent systems provide a convenient framework for the massive deployment of agents based on Large Language Models (LLMs). It provides a sound framework where agents can interact with existing software to perform actions in an autonomous way. Agentic AI is the new term to denote these AI-based autonomous systems. In real multi-agent systems, agents need to deal with malicious agents some of which can exploit security vulnerabilities to do harm. A memory poisoning attack is an example of exploiting a vulnerability to modify agents knowledge and, thus, affect its behavior. In an agent, there are different types of memory systems (semantic, episodic, and short-term) which have their own characteristics and for which different memory poisoning attacks may be developed. In this paper we focus on semantic memory, we discuss its attack, and mitigation strategies. We introduce private knowledge retrieval and private inference. We propose a signature system in the knowledge base to prevent from undesired manipulation. This should avoid memory poisoning attacks by introducing security by design in the construction of the semantic memory. In our architecture, an agent contains a local Prolog-like inference system (i.e., using Horn clauses to represent knowledge) which accesses an external knowledge base to get the required knowledge. This externally signed knowledge base is accessed privately. We provide different implementations in Python using different privacy levels including k-anonymity and information-theoretical privacy (IT-PIR).
Abstract (translated)
URL
https://arxiv.org/abs/2611.13696
PDF
-
Walrus Technical Report
2026-11-10 07:37:38Alkis Koudounas, Osamu Take, Hayato Futami, Quentin Jodelet, Emiru TsunooAbstract
We present Walrus, a unified 8B speech-and-audio foundation model handling speech recognition (including code-switching and multi-talker speech), speech-to-text and speech-to-speech translation, text-to-speech, emotion recognition, audio captioning, and general audio understanding. Our main contribution is not a new architecture but a data-centric training recipe; we use a standard decoder-only backbone initialized from Gemma 4, so that gains can be traced to the data. For speech generation, Walrus predicts low-bitrate discrete speech tokens, and a lightweight token-to-waveform decoder converts them into high-fidelity audio. The recipe is a three-stage data curriculum that moves from broad multilingual data to multi-talker, code-switching, and domain-specific data. At each stage, task-aware and dataset-aware weighting balances tasks with very different token budgets and protects core skills from forgetting. Trained on over 600M samples covering more than 20 tasks and 100 languages, Walrus leads on most multilingual, multi-talker, and code-switching ASR, speech translation, and audio understanding benchmarks, outperforming both open speech language models of comparable size and dedicated task-specific models. On most benchmarks, it even matches or surpasses the $4\times$ larger Qwen3-Omni's performance. Results suggest that data engineering is a primary driver of quality in multi-task speech and audio modeling.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13689
PDF
-
You Are What You Read: Indian Political News Media Bias Dataset for Hindi Language
2026-11-10 07:34:05Kushal Trivedi, Mayank Mahor, Dhruv Sanghvi, Roshni ChakrabortyAbstract
Media bias understanding and detection remains an under-researched problem for Indian context, specifically for Indic languages, such as, Hindi, which is widely spoken in India. In this paper, we introduce a comprehensive, multi-annotator, explainable dataset consisting of 448 unique articles from four leading Indian news agencies publishing in Hindi. We focus on Hindi, the most spoken and an official language of India. The study of media bias is particularly relevant in the Indian context due to the country's linguistic diversity and large population, where the same event is often reported by multiple news agencies in different languages, thereby introducing scope for bias. The dataset comprises Hindi political news articles annotated by 14 experts and classified into four bias categories - selection bias, framing bias, clickbait bias and omission bias - when an article is identified as biased or unbiased. For each biased article, annotators also provide a textual explanation, enabling future semi-automated approaches to incorporate explainability when language models reason over the dataset. Cohen's Kappa scores were 0.81 for simple bias/no-bias classification reflecting strong inter-annotator agreement. For the multi-label task, the Cohen's Kappa values are 0.89, 0.75, 0.81, and 0.79 for clickbait, selection, framing, and omission biases, respectively, resulting in an average Kappa of 0.80, which also reflects strong agreement.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13686
PDF
-
AURA: Auditory Spatial Intelligence via a Unified Representation from a General-Audio Encoder through Complete Alignment
2026-11-10 07:34:05Boyan Han, Xiaoyan Wang, Zhe Sun, Ziqi Wang, Yujun Cai, Hongli Cao, Yiwei Wang, Ming-Hsuan Yang, Liang AnAbstract
Answering spatial questions about multi-source scenes requires a model both to identify each sound and to locate it. Existing spatial audio-language models obtain spatial information from a dedicated encoder, which separates spatial and semantic processing into distinct feature spaces. We study whether a general-audio codec trained without spatial supervision can serve both purposes. A controlled layer-wise analysis shows that the frozen codec retains spatial information, with lateral cues carried by the difference between the two ear channels. This information, however, is not readily accessible: in dual-source scenes, a model trained on these features predicts left/right at chance and, on the remaining axes, assigns the detected direction to the queried and the non-queried source equally often. We attribute this failure to an audio--text information asymmetry, in which the audio contains both sources while the answer describes only the queried one. A controlled $2\times2$ experiment shows that either completing the scene description or exposing the inter-channel difference lifts left/right accuracy above chance, whereas binding directions to the queried source requires both. Building on these findings, AURA keeps the codec and the language model frozen and exceeds the strongest dedicated system on SpatialSoundQA by 10.31 points in dual-source direction accuracy, with about 40\% of its trainable parameters.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13687
PDF
-
Bootstrapped On-Policy Distillation
2026-11-10 07:30:34Yue Wu, Xueyu Zhou, Yangrong Hu, Jian HuangAbstract
On-policy distillation (OPD) improves language models by distilling teacher predictions on student-generated trajectories, while on-policy self-distillation (OPSD) removes the need for an external teacher by using a reference-conditioned copy of the model as the teacher. Despite its great early improvements, we find that OPSD does not remain consistently effective under continued training and can eventually suffer substantial performance degradation. We study this training behavior and link the degradation to a growing mismatch between the evolving reference-free student and the frozen privileged teacher. In the early stage of OPSD, the privileged teacher remains closely aligned with the student and provides useful guidance for improving the student's generation trajectories. However, as training progresses, the growing distributional mismatch between the privileged teacher and reference-free student can cause the teacher to provide misleading supervision, ultimately degrading student performance. This phenomenon motivates us to first leverage privileged reference supervision to acquire useful knowledge and then transfer it to a fresh student. We introduce Bootstrapped On-Policy Distillation (BOPD), a two-stage framework that first uses OPSD to internalize task-relevant information from reference-conditioned supervision into an intermediate model, and then distills this reference-free intermediate model into a fresh copy of the initial model. Experiments across models show that BOPD improves mathematical reasoning and code generation beyond both the initial models and the constructed OPSD teachers. These results suggest that privileged reference supervision can be more effectively exploited by first transferring its task-relevant signal into a task-adapted model and then distilling from that model without reference conditioning.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13680
PDF
-
OpticalRAG: Pixel-Space Compression for Token-Efficient Retrieval-Augmented Generation
2026-11-10 07:24:50Senhao Liu, Yuheng Zhang, Chunyu Wei, Yueguo Chen, Xinran ZhangAbstract
Retrieval-augmented generation (RAG) is bottlenecked by the token budget of large language models: feeding long documents into the context window is expensive, while aggressive textual compression discards fine-grained information that is irreversibly lost in a one-dimensional token sequence. We argue that this bottleneck is not fundamental to information density but to the choice of compression \emph{modality}. Modern visual encoders can pack the contents of a rendered text page into roughly a hundred visual tokens with little semantic loss, opening a new compression-fidelity tradeoff that purely textual methods cannot reach. We present \textbf{OpticalRAG}, a framework that renders documents as page images, encodes them with a frozen vision encoder, and distills the retrieved visual tokens into as few as $16$ injected tokens before passing them to the LLM. Two ingredients make this work without expensive cross-modal training: (i) \emph{Encoder-Bridged Retrieval}, which projects text-query states into the visual feature space of the frozen encoder, through a projection shared with the compressor, to retrieve visual tokens from a text query, with no contrastive training; and (ii) \emph{Query-Driven Distillation}, a lightweight cross-attention compressor that condenses the retrieved visual tokens into a small set of query-conditioned prefix tokens. On LongBench-v2, BAMBOO, and LooGLE-v2, OpticalRAG achieves the best compression-aware accuracy (CAP) on all three benchmarks while injecting only $16$ tokens, and improves the closed-book Qwen2.5-7B-Instruct backbone by $5.17$ points on this http URL is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13675
PDF
-
A computational framework for understanding embodied plans
2026-11-10 07:15:38Aalap D. Shah, Basil Saeed, Joshua B. Tenenbaum, Tobias Gerstenberg, Hyowon Gweon, Ilker YildirimAbstract
Reasoning about people's plans and goals based on their actions is fundamental to human social interaction and essential for cooperative, embodied artificial intelligence. Formalizing this capacity in naturalistic, embodied settings has been a longstanding challenge. Here, we present GenEA, a novel agent representation capable of reasoning over multi-step, long-horizon embodied plans. At the heart of GenEA lies a multilevel state-space that couples the coarse biomechanics of an articulated body and its interactions with nearby objects by extending the task-and-motion planning framework from robotics with simulation-based intuitive physics. By embedding this state-space in a generative model, GenEA yields a joint probability distribution over goals and embodied plans. Across several experiments spanning the qualitatively distinct domains of object manipulation and navigation, featuring diverse tasks and multiple behavioral measures, we show that GenEA, through simple transformations of this joint distribution, recapitulates how humans reason about embodied plans. Crucially, GenEA consistently outperforms strong alternative models, including plausible heuristics, standard task-and-motion planning, and a state-of-the-art vision-language model. GenEA thereby offers a new perspective on the architecture of human reasoning and social cognition, suggesting that embodied artificial intelligence systems seeking to seamlessly interact with humans should adopt GenEA-like generative models of action planning.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13669
PDF
-
MNN-Pocket-0.3B Technical Report
2026-11-10 06:53:47MNN TeamAbstract
This technical report presents MNN-Pocket-0.3B, a compact 300M-parameter dense language model designed for efficient on-device deployment and supporting context lengths of up to 32K tokens. With an INT4 deployment footprint of only 173 MB and decoding throughput of up to 466 tokens/s on iPad Pro M5 (Metal), it can be seamlessly integrated into applications on edge devices such as smartphones and smart glasses. MNN-Pocket-0.3B is developed through the co-design of its architecture, data, training pipeline, and inference system. Hardware-aware architecture selection and dependency-aware vocabulary pruning balance model capability with on-device memory and compute constraints. A four-stage pre-training pipeline covering approximately 4.1 trillion high-quality Chinese and English tokens equips the model with broad knowledge and reasoning capabilities. This is followed by a three-stage post-training pipeline comprising instruction fine-tuning, multi-teacher on-policy distillation, and quantization-aware training. Finally, hardware-specific optimizations in the MNN inference engine enable fast, memory-efficient inference and long-context processing. Extensive evaluations demonstrate that MNN-Pocket-0.3B achieves frontier-level performance among models of comparable size, with particularly strong results in Chinese and English knowledge and reasoning, mathematics, coding, and instruction following. It consistently outperforms similarly sized open-source models and even surpasses several 0.5B-parameter models on instruction-following benchmarks. We release MNN-Pocket-0.3B with open weights and ready-to-use MNN deployment packages. We hope it will provide a practical foundation for capable, responsive, and memory-efficient edge applications while advancing research and development in on-device language models.
Abstract (translated)
URL
https://arxiv.org/abs/2611.12311
PDF
-
DrugCoding: A Scientific Programming Agent for Drug Discovery
2026-11-10 06:47:15Junwei Hu, Zhuohao Lin, Kun Li, Hengnan ZhangAbstract
Drug discovery involves multiple interconnected research stages, including molecular design, property prediction, and virtual screening. These processes are characterized by complex workflows, interdependent computational tasks, and diverse specialized tools, placing substantial demands on the coordination of scientific models and computational resources. However, existing large language model (LLM)-based scientific agents still face challenges in autonomous planning of complex tasks, adaptive selection of specialized models, and reliable execution of multi-stage scientific computations, limiting their effectiveness in automating complex drug discovery workflows. To address these challenges, we propose DrugCoding, a scientific programming agent system that integrates scientific task planning, adaptive model routing, multi-agent collaboration, and programmatic tool execution. DrugCoding enables autonomous task planning and dynamic model adaptation for complex research objectives. It can flexibly invoke specialized scientific models, autonomously generate and execute code according to task requirements, and leverage multi-agent collaboration to support efficient execution and reliable validation of complex scientific tasks. To evaluate DrugCoding's performance in scientific tool invocation and multi-stage task execution, we conduct multi-turn tool-calling experiments across diverse language models and scientific tool interfaces. Experimental results demonstrate that DrugCoding achieves an average session-level Pass@1 of 97.35% on the full benchmark, outperforming the strongest baseline, Codex, by 9.24%. On the challenging subset, DrugCoding achieves 97.92%, exceeding the strongest baseline.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13651
PDF
-
MOAI: Module-Optimizing Architecture for non-interactive secure transformer Inference
2026-11-10 06:36:43Linru Zhang, Xiangning Wang, Jun Jie Sim, Zhicong Huang, Jiahao Zhong, Huaxiong Wang, Pu Duan, Kwok-Yan LamAbstract
Privacy concerns have been raised in Large Language Models (LLM) inference when models are deployed in Cloud Service Providers (CSP). Homomorphic encryption (HE) offers a promising solution by enabling secure inference directly over encrypted inputs. However, the high computational overhead of HE remains a major bottleneck. To address this challenge, we propose MOAI, an efficient HE-based, non-interactive framework for secure transformer inference. MOAI gains significant efficiency improvement from: (1) a novel evaluation flow that combines column and diagonal packing with consistent strategies across all layers, eliminating expensive format conversions. (2) rotation-free algorithms for Softmax and LayerNorm that significantly reduce the number of costly HE rotations, removing 2,448 HE rotations in BERT-base inference. (3) Column packing removes rotations in plaintext-ciphertext matrix multiplications and interleaved batching further reduces the rotations in ciphertext-ciphertext matrix multiplications. MOAI uses at least 1.7x fewer HE rotations compared to the state-of-the-art works across all matrix multiplications of BERT-base. As a result, we achieve a $52.8 \%$ reduction in evaluation time compared to the state-of-the-art HE-based non-interactive secure transformer inference, THOR (Moon et al., CCS 25). We then apply MOAI on the Powerformer's framework and achieve a $55.7 \%$ reduction in evaluation time compared to Powerformer (Park et al., ACL 25), which approximates Softmax and LayerNorm with simpler functions in transformer and proposes HE-based non-interactive transformer inference. We report an amortized time of 2.36 minutes per input on a single GPU environment. We show the extendibility by applying MOAI in LLaMA-3-8B. Our implementation is publicly available as open source.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13641
PDF
-
Synapse-Control: Making Agent Experience Actionable with Lightweight Runtime Control
2026-11-10 05:50:12Zhiyuan Huang, Shengan Zheng, Chenghu Yang, Song Zhang, Lingyun Li, Chen Yu, Weiquan Huang, Linpeng HuangAbstract
Large language model agents reuse execution experience to improve reliability on complex tool-based tasks. Systems such as Sentry call large models to generate failure diagnoses before retrieving advice, adding inference cost to experience reuse. We present Synapse-Control, an agent runtime whose lightweight control plane selects advice and determines when to send it to the task-executing model. Rules shortlist relevant advice, and a lightweight 1.7B classifier fine-tuned on teacher-labeled state-suggestion pairs scores those candidates instead of generating a diagnosis. Micro-contracts encode reusable interface checks, such as required arguments or valid options, without prescribing complete task programs. A Prompt Value Table records earlier guidance and feedback to limit repeated suggestions through cooldown and token-budget checks. Across 1,308 attempts per method in AppWorld and WebShop, Synapse-Control achieves 270 successes (29.8% more than Base), replaces 4,987 generative diagnosis calls, and reduces auxiliary control time relative to Sentry by 50.2% and 51.0%, respectively.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13615
PDF
-
When to Trust Adaptation: Contextual Fusion for Transferable Vision-Language Prompting
2026-11-10 05:28:38Zhenyuan Chen, Wenshuo Tao, Zhuo Huang, Shuo Chen, Ming-Ming Cheng, Xiang LiAbstract
When a vision-language model encounters a new class or domain, how much should it trust its learned prompts? Because source adaptation can correct pretrained errors on some target images while disturbing valid predictions on others, we introduce Contextual Adaptive Fusion (CAF), which retains four prediction branches by pairing frozen and prompted visual and textual representations. To arbitrate when branches disagree on an image, CAF combines each branch's predictive confidence with how frequently the branches favor its proposed class across the unlabeled target collection, placing greater trust in predictions that are locally confident yet less dominated by collection-wide frequency bias. The fusion coefficients are learned once on source data and remain fixed at inference. Trained on ImageNet, CAF achieves 69.96% average accuracy across ten cross-dataset benchmarks (1.92 percentage points above ManiPT) and remains highly competitive across four ImageNet variants, setting a new best of 52.21% on ImageNet-S. In base-to-novel transfer across eleven datasets, CAF further reaches 79.77% novel-class accuracy and an 82.81% harmonic mean. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2611.11824
PDF
-
Towards Reliable VLM Judges: State-Conditional Invariance and Presentation-Aware Diagnostics
2026-11-10 05:15:50Qinan Zhang, Xinyu Wang, Qihang Jin, Tianyuan Wang, Hongbing Luo, Pengfei Li, Wenjun Lv, Yu Kang, Jian DiAbstract
Vision-language models (VLMs) are increasingly used as automated judges for multimodal and embodied tasks, yet aggregate accuracy or consistency alone does not reveal whether a judge changes its verdict for the right reason. We propose State-Conditional Invariance (SCI), a reliability principle requiring a judge to remain stable under semantics-preserving changes in presentation while responding to causal interventions that change the task outcome. We instantiate SCI in GridWM-Judge, a simulator-grounded diagnostic benchmark built from deterministic MiniGrid trajectories with aligned Full, evidence-removal, and counterfactual variants, plus presentation and representation probes. Across five diagnostic tasks, GridWM-Judge tests local perception, action-conditioned transitions, presentation stability, representation exchangeability, and Full-CF causal discrimination. Our diagnostic evaluations reveal heterogeneous failure modes, including sensitivity to evidence presentation and encoding, low activation on successful trajectories, and weak paired counterfactual discrimination. Perception-controlled analyses further indicate that visual grounding and objective recovery are substantial, model-dependent bottlenecks. These results motivate jointly reporting correctness, presentation stability, and causal sensitivity instead of relying on a single score.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13596
PDF
-
IFRS Compliance Verification: An Open Multi-LLM Benchmark and Content-Conditioned Ensemble Classifier
2026-11-10 05:12:13Aghoghomena Akasukpe, Li Yang, Xinyao Zhou, Chen Shen, Miguel Vargas MartinAbstract
Verifying corporate financial disclosures against the International Financial Reporting Standards (IFRS) is a core auditing task in which experts cross-reference disclosures with hundreds of fine-grained regulatory paragraphs: a laborious, costly, and error-prone process that motivates growing interest in large language models (LLMs) as compliance assistants. Yet single-model accuracy remains insufficient for high-stakes use, and naive ensembling is vulnerable to systematic per-model biases that can drag aggregate accuracy below that of the strongest individual annotator. This gap is addressed with a multi-stage framework, supported by a new benchmark of 19,998 IFRS regulation-disclosure pairs comprising 11,161 unanimous multi-LLM labels and 8,837 human annotations on items where the LLMs disagreed. The framework couples a three-stage domain-adapted sentence embedding pipeline (BERT masked language modeling on SEC EDGAR filings, unsupervised TSDAE and SimCSE adaptation, and supervised contrastive alignment) with a learned ensemble classifier that fuses frozen embeddings and LLM probability features from GPT-4.1, Claude Sonnet 4, and Gemini 2.5 Pro via Tree-structured Parzen Estimator (TPE) optimized gradient-boosted trees. On the disagreement set, the ensemble achieves 81.7% weighted F1, exceeding the strongest single LLM by 10.4 points and majority voting by 14.6 points, and outperforms eleven unsupervised aggregation baselines; the retrieval pipeline attains Recall@15 of 45.3% on unseen regulations. The dataset, model checkpoints, annotation platform, and classifier implementation will be publicly released.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13595
PDF
-
Transformers Provably Learn Graph Search: Training Dynamics and the Exponential Power of Depth
2026-11-10 05:09:36Xutao Ma, Somayeh SojoudiAbstract
Large Language Models (LLMs) have achieved remarkable progress on complex reasoning tasks, yet the mechanisms by which they acquire reasoning capabilities and the efficiency of their reasoning remain poorly understood. In this paper, we investigate the path-finding problem over directed graphs, a symbolic abstraction of multi-step reasoning. We first characterize the architectural requirements for this task, proving that one-layer transformers require $\Omega(N^2)$ size on dense graphs to implement the key DFS child-selection primitive. While a two-layer transformer implements full DFS with $O(N\log(N))$ size. Thus, a second layer is necessary and sufficient for near-linear DFS reasoning. We then extend this expressivity analysis to deeper models: for every $k\geq3$, a $k$-layer transformer implements a bounded-reachability DFS policy with look-ahead horizon $1+2^{k-3}$, establishing an exponential increase in reachability horizon with depth. Finally, going beyond expressivity, given appropriately curated training data, we show that two-layer transformers can provably learn to execute DFS via gradient flow and generalize to unseen graphs. Together, our results provide a mechanistic account of how transformers implement graph search, how depth expands bounded look-ahead, and how such reasoning algorithms can emerge through training.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13594
PDF
-
Learning Perturbation Robust Policies for LLM Agents with Stable Optimization
2026-11-10 05:09:29Pengxin Wang, Yuanzhe LI, Yuxin Ren, Huanrui Yang, Jingdi ChenAbstract
Reinforcement learning (RL) has become an effective post-training paradigm for long-horizon large language model (LLM) agents. However, we find that the resulting policies can be sensitive to various policy perturbations, such as hidden-state noise, pruning, and quantization. In this work, we study how to improve perturbation robustness during policy optimization. We first introduce the notion of a perturbation robust policy and analyze conditions under which perturbed policy updates preserve stable monotonic improvement. Based on this analysis, we introduce Stable Perturbation-Robust Policy Optimization (SPrPO), which applies adaptive and sensitivity-aware perturbations during RL training. We evaluate SPrPO on ALFWorld and WebShop and conduct systematic experiments across multiple perturbation types and scales, showing improved perturbation robustness while maintaining stable policy optimization.
Abstract (translated)
URL
https://arxiv.org/abs/2609.34064
PDF
-
POEM: Output-Relative Block-Sparse Attention for Long-Context Prefilling
2026-11-10 04:52:58Haoyi Wu, Yunkai Dang, Wanyu Zhang, Lei Wang, Wenbin LiAbstract
Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for reliable deployment. Jailbreak attacks use diverse forms of adversarial prompting to bypass alignment safeguards and induce LLMs to generate harmful responses. Some representation-based defenses regulate internal activations using predefined steering directions or learned safety regions. However, adapting intervention to each input without a predefined representational target while limiting over-refusal remains a challenge. In this paper, we propose Target-free Entropy-guided Residual Modulation (TERM), an inference-time defense that uses prompt-calibrated attention entropy deviations to guide adaptive modulation of the model's own residual updates. Attention patterns vary across prompts and decoding steps, making raw entropy values difficult to interpret as a consistent signal for intervention. We therefore calibrate attention entropy against each prompt's own prefill baseline to establish an input-specific reference for reducing unnecessary intervention on benign requests. Adaptive scale normalization and temporal accumulation then account for differences in deviation scale and persistence to determine intervention strength. The resulting signal controls bounded scaling of the model's own residual updates, without relying on a predefined safety direction. We compare our method with seven defenses on seven jailbreak attacks and two over-refusal benchmarks across Llama, Qwen, and Gemma models, including Qwen3 models from 0.6B to 32B. Results show our method improves jailbreak robustness while maintaining low over-refusal on benign safety-sensitive requests with modest inference overhead.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13578
PDF
-
TERM: Target-Free Jailbreak Defense via Entropy-Guided Residual Modulation
2026-11-10 04:52:10Yunkai Dang, Wanyu Zhang, Meiyi Zhu, Qi Fan, Tianyu Ding, Wenbin LiAbstract
Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for reliable deployment. Jailbreak attacks use diverse forms of adversarial prompting to bypass alignment safeguards and induce LLMs to generate harmful responses. Some representation-based defenses regulate internal activations using predefined steering directions or learned safety regions. However, adapting intervention to each input without a predefined representational target while limiting over-refusal remains a challenge. In this paper, we propose Target-free Entropy-guided Residual Modulation (TERM), an inference-time defense that uses prompt-calibrated attention entropy deviations to guide adaptive modulation of the model's own residual updates. Attention patterns vary across prompts and decoding steps, making raw entropy values difficult to interpret as a consistent signal for intervention. We therefore calibrate attention entropy against each prompt's own prefill baseline to establish an input-specific reference for reducing unnecessary intervention on benign requests. Adaptive scale normalization and temporal accumulation then account for differences in deviation scale and persistence to determine intervention strength. The resulting signal controls bounded scaling of the model's own residual updates, without relying on a predefined safety direction. We compare our method with seven defenses on seven jailbreak attacks and two over-refusal benchmarks across Llama, Qwen, and Gemma models, including Qwen3 models from 0.6B to 32B. Results show our method improves jailbreak robustness while maintaining low over-refusal on benign safety-sensitive requests with modest inference overhead.
Abstract (translated)
URL
https://arxiv.org/abs/2611.13577
PDF
