Multimodal agents use web and image tools to answer questions beyond their internal knowledge. Some search tasks use images mainly to identify an initial entity, then rely on text retrieval. These tasks offer limited insight into how agents connect information across images, objects, and search steps. We therefore focus on visual context management. This involves organizing and using visual information and linking it to other findings during search. Visually intensive tasks require agents to combine object appearance, local details, or spatial relations with external information. We introduce VisSearch-Bench, with 204 human-written tasks across 7 main domains and 25 subdomains. It covers Set-wise Visual Search, Sequential Visual Search, Hybrid Wide-Deep Search, Competitive Visual Disambiguation, and Visual-Structure-Guided Search. Each task provides a reference answer, traceable evidence, and scoring rules. Accuracy measures complete task success, while Process Score measures the correctness and completeness of individual results and subquestion answers. Across 16 multimodal models, the best system fully solves only 26.47% of tasks, even with search tools. These results highlight the challenge of linking visual evidence with search results to produce complete answers.
Direct Preference Optimization (DPO) minimizes empirical risk over observed preference comparisons, implicitly assuming that the training preference distribution matches the deployment preference distribution. In practice, this distributional assumption often fails, as preference distribution shifts can arise at three levels: group-level shift, prompt-level shift, and preference-level shift. Such multi-level distributional mismatch poses a critical challenge for safe LLM deployment, yet existing robust DPO methods typically either model uncertainty over the joint preference distribution or focus on individual sources of shift, leaving the hierarchical structure of OOD preference shifts underexplored. We propose TPO, a tri-level distributionally robust framework for OOD preference alignment, which jointly models all three levels of distributional uncertainty through a hierarchically structured rectangular ambiguity set. By instantiating it with Wasserstein distance, we derive a tractable and interpretable objective that combines a preference-level margin penalty, prompt-level adversarial embedding perturbation, and worst-case group reweighting, each governed by an independent radius. We establish finite-sample convergence guarantees and empirically validate the effectiveness of TPO across diverse OOD environments.
Passenger electroencephalography (EEG) can reveal road risk through anticipatory responses before hazardous events and neural reactions around their onset. However, existing EEG decoding methods often map physiologically meaningful arrangement of electrodes onto artificial 2D grids and thus flatten the higher-order time-feature representations into matrices, obscuring meaningful structure in the data. To preserve and exploit the spatial and time-feature structure of EEG recordings, we propose the Graph-Tensor EEG Decoding Model (GraTED) for passenger risk prediction (RP) and danger identification (DI) in autonomous-vehicle scenarios. GraTED models the electrode arrangement as an undirected graph based on their scalp locations, using graph convolution to capture spatial dependencies and low-rank tensor classification to preserve the multidimensional time-feature structure, while controlling model complexity. Evaluated on the PassengerEEG dataset, comprising 11 participants and 45 hours of recordings, GraTED achieves the highest balanced accuracy, compared to the state-of-the-art, across all three evaluation settings, while requiring the fewest learnable parameters among seven baseline methods.
We study post-training for dynamically routed mixture-of-experts (MoE) models, focusing on integrating independently optimized specialists while keeping consolidation efficient. Our approach combines calibrated multi-teacher on-policy distillation (C-MOPD), which matches each selected teacher on a local Top-$K$ support with teacher-wise loss normalization and clipping, with an execution path that uses post-capacity bounded-width dispatch, final-route replay, and asynchronous rollout--training overlap. On Turing-20B-A2B, the selected model achieves a four-domain normalized integration score of 0.9988 and retains at least 90.16\% of the measured specialist gain in every domain. At matched $K=32$, the full calibration objective scores 0.9962, compared with 0.9724 without calibration and 0.9834 without clipping. At a matched 8K response cap, the model exceeds both shared SFT and Qwen3-8B on all four reasoning-enabled mathematics and code benchmarks (unweighted mean: 85.92, versus 80.33 for Qwen3-8B and 74.21 for SFT), and improves over SFT on ten of twelve broader evaluation metrics. In a PPU-based profile, the combined execution configuration reduces consolidation step time from 398.7 to 139.3 seconds, an observed 2.86$\times$ speedup over serial execution. These results demonstrate strong specialist integration alongside more efficient consolidation, while reflecting the complete training and execution framework rather than any single component.
David Hang, Wenkai Yang, Kuiye Ding, Haiyang Xin, Jacky Wei
Standardized benchmarks are fundamental to reliable progress in AI for EDA, including learning-based thermal modeling. However, existing thermal prediction studies often rely on different datasets, simulators, data splits, preprocessing pipelines, and metrics, while most datasets and implementations remain unavailable, making fair and reproducible comparison difficult. We introduce IC-ThermBench, an open and progressive benchmark that combines established 3D-IC steady-state, transient, and industrial package tasks with a new 50,000-sample 2.5D chiplet extension designed to evaluate progressively broader represented physical variation and cross-package OOD transfer. Five Generalization Scopes cover 3D-IC fixed-design prediction, Within-Family Generalization under layout, material, and boundary-condition variation, and Cross-Package OOD transfer to unseen package systems. We evaluate eight representative baselines under common data, splits, labels, and metrics. Performance degrades gradually from S2 to S4 as represented physical support broadens, but Cross-Package OOD produces a much sharper degradation: the best RMSE and MAE increase from 1.216 and 0.938~K at S4 to 15.99 and 15.00~K at S5, respectively. With only 10 labeled samples per OOD case, target-domain adaptation reduces the best MAE to 2.60~K. IC-ThermBench further provides a unified generation, training, inference, and evaluation pipeline, enabling reproducible and fair comparison of existing and new thermal predictor.
Many uses of language models end in a typed decision: a choice among a closed set of options, whose probabilities code can act on. Reading those probabilities from one forward pass is fast but often wrong on decisions that take several steps, such as arithmetic, counting, and date comparison. Letting the model think first fixes most of these errors, but costs seconds per decision instead of milliseconds. To pay for thinking only where it helps, ThinkJev builds a fast and a slow system from one open hybrid-thinking model that can answer directly or after a thought. System 1 reads the model's probabilities for the option letters when it answers directly; System 2 reads the same letters after the model has thought on the same cached prompt, so escalating repeats no work and both systems give probabilities over the same options. Because the letters can be read after any number of thought tokens, one sampled thought also gives the answer under every thinking budget. Without training, ThinkJev lets System 1 answer alone only where a statistical test on labeled dev data guarantees that its accepted answers stay under a target error rate. At a 5% target, Qwen3.5-27B answers 24.5% of 8359 test decisions this way with a realized error of 3.9% (Qwen3.5-4B and 9B: 6.5% and 0.0%). For the rest, stopping each thought as soon as its answer is confident reaches 81.6% accuracy, against 82.7% for always thinking, with 72% of the thought tokens. Choosing when to think from System 1's confidence alone saves less, because the decisions System 1 is least sure of are often not the ones thinking fixes. On the public items of JevBench, an independent benchmark of typed-decision services, Qwen3.5-27B thinks on 29.4% and is not distinguishable in accuracy from the two thinking APIs listed (exploratory). Code and data generators are available at https://github.com/ding-xiaolu/ThinkJev.
Open-Set Test-Time Adaptation (OSTTA) operates on non-stationary target streams where covariate and semantic shifts coexist. Existing TTA methods often obtain adaptation signals by updating model parameters or maintaining mutable memories, which can be costly and vulnerable to unknown samples. We present TALOR, a lightweight, backward-free rectification module that uses the source linear head as a geometric anchor. The head-induced basis decomposes each normalized feature into high-gain principal coordinates and low-gain tail coordinates. TALOR estimates a soft-routed affine correction in the principal subspace, combining a regime-level principal bias with a tail-conditioned slope that captures sample-level tail-to-principal coupling. Using weighted ridge regression, it subtracts this drift estimate from the principal coordinates, reconstructs the feature and feeds it into the source head. On the ImageNet-C benchmark with six csOOD datasets, TALOR outperforms the next-best UniEnt by 2.2% H-score while running 2.5x faster and using only 12% of its GPU memory; as a post-correction plug-in, it further boosts COME to 68.4% H-score with only 2% additional memory overhead.
Guirong Chen Renmin University of China, Shuqi Ye, Wenkai Yang Renmin University of China, Shiqi Shen Wechat, Tencent, Guangyao Shen Tencent, Yankai Lin Renmin University of China
The evolution paradigm of Large Language Models (LLMs) is shifting from scaling training compute to scaling inference-time compute. While Reinforcement Learning with Verifiable Rewards (RLVR) has become a key engine for this transition, standard approaches often fail to equip models with the autonomous improvement capabilities required for test-time scaling. Existing critique-guided methods attempt to mitigate this by leveraging external feedback or ground-truth signals; however, these dependencies are unavailable at test time, fundamentally limiting the model’s capacity for continuous self-improvement. To bridge this gap, we propose CURE (Critique-driven Unified REinforcement Learning), a framework that jointly optimizes a single policy for standard solving, critiquing, and guided re-exploration. Uniquely, CURE facilitates re-exploration by generating strategic hints while discarding initial incorrect solutions to mitigate anchoring bias.Empirical results across diverse mathematical reasoning and code generation benchmarks demonstrate that CURE not only maintains competitive single-turn performance but, more importantly, unlocks effective inference-time scaling, enabling the model to significantly boost accuracy through iterative self-improvement.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhihao Shuai, Yiyun Chen The Hong Kong University of Science and Technology, Maolin Ma, Yutong Chen, Hanjia Qiu, Jing Xu The Hong Kong University of Science and Technology (Guanzhou), Ziye Chen, Weikai Yang Hong Kong University of Science and Technology (Guangzhou)
Natural Language to Excel Formula (NL2Formula) translates user intent into executable spreadsheet formulas. However, current models often produce near-miss outputs—formulas that parse correctly yet fail at execution due to an incorrect function, operator, or reference. Through a systematic error analysis, we find that these errors repeatedly arise from a small set of structural decision points, motivating the need for typed error supervision rather than general error signals. To this end, we introduce an abstract syntax tree (AST)-based error taxonomy that organizes common error modes by the kind of decision that goes wrong in the parse tree. Building on this taxonomy, we propose Error-Aware Contrastive Few-Shot Learning (ECFL), an error-aware framework that unifies training and inference around typed error supervision. During offline training, ECFL mines near-miss errors, assigns error types under the taxonomy, and builds error-aware contrastive demonstrations for fine-tuning. During online inference, a lightweight predictor estimates likely error types and triggers targeted retrieval of contrastive demonstrations to guide single-pass decoding. Experiments show ECFL improves Exact Match (EM) by 6.4 points over supervised fine-tuning (SFT) and matches self-consistency (SC@5) accuracy at substantially lower inference cost.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhiyuan Peng, Xin Yin, Pu Zhao, Fangkai Yang, Lu Wang Microsoft, Ran Jia Microsoft, Xu Chen, Qingwei Lin Microsoft Research 等 10 位作者
Large language models and agents have achieved remarkable progress in code generation. However, existing benchmarks focus on isolated function/class-level generation (e.g., ClassEval) or modifications to existing codebases (e.g., SWE-Bench), neglecting complete microservice repository generation that reflects real-world 0-to-1 development workflows. To bridge this gap, we introduce RepoGenesis, the first multilingual benchmark for repository-level end-to-end web microservice generation, comprising 106 repositories (60 Python, 46 Java) across 18 domains and 11 frameworks, with 1,258 API endpoints and 2,335 test cases verified through a “review-rebuttal” quality assurance process. We evaluate open-source agents (e.g., DeepCode) and commercial IDEs (e.g., Cursor) using Pass@1, API Coverage (AC), and Deployment Success Rate (DSR). Results reveal that despite high AC (up to 73.91%) and DSR (up to 100%), the best-performing system achieves only 23.67% Pass@1 on Python and 21.45% on Java, exposing deficiencies in architectural coherence, dependency management, and cross-file consistency. Notably, RepoGenesis-8B, fine-tuned on RepoGenesis (train), achieves performance comparable to GPT-5 mini, demonstrating the quality of RepoGenesis for advancing microservice generation. We release our benchmark at https://github.com/pzy2000/RepoGenesis.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Ruiyao Xu Northwestern University, Mihir Parmar Google, Tiankai Yang University of Southern California, Zhengyu Hu University of Washington, Yue Zhao University of Southern California, Kaize Ding Northwestern University
Learning from preference-based feedback has become an effective approach for aligning LLMs across diverse tasks. However, high-quality human-annotated preference data remains expensive and scarce. Existing methods address this challenge through either self-rewarding, which scales by using purely AI-generated labels but risks unreliability, or active learning, which ensures quality through oracle annotation but cannot fully leverage unlabeled data. In this paper, we present CoAct, a novel framework that synergistically combines self-rewarding and active learning through strategic human-AI collaboration. CoAct leverages self-consistency to identify both reliable self-labeled data and samples requiring oracle verification. Additionally, oracle feedback guides the model to generate new instructions within its solvable capability. Evaluated on three reasoning benchmarks across two model families, CoAct achieves average improvements of +13.25% on GSM8K, +8.19% on MATH, and +13.16% on WebInstruct, consistently outperforming all baselines.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表