Under expert parallelism, mixture-of-experts (MoE) layers pay all-to-all token dispatch, grouped GEMMs, and a combine step in every layer. In decode each rank holds only 1-256 tokens: dispatch latency and weight traffic dominate while tensor-core utilization is inherently low. In fused operators such as DeepGEMM's Mega MoE, dispatch and computation overlap only by sharing every SM: dispatch warps compete with TMA/MMA and epilogue warps for registers, shared memory, and issue slots; every phase shares one CTA resource budget, synchronization spans the whole SM array, and every replay ends with a workspace-clean pass and a cross-rank barrier. We present GC-MoE, an open-source FP8/FP4 MoE inference operator for NVIDIA Hopper and Blackwell GPUs that uses CUDA Green Contexts to split the device into a small communication partition and a large compute partition. Each role gets its own kernel, resource budget, and synchronization domain, so dispatch and GEMM tune independently. A three-node CUDA Graph (four on Hopper with fused shared experts) launches the dispatch and grouped-GEMM kernels as concurrent roots; a token-ring handshake starts computation on the first filled ring blocks, and double-buffered receive counters remove the trailing cross-rank barrier between replays. The compute partition runs operand-swapped tcgen05 block-scaled MMA on Blackwell and register-A WGMMA with in-register FP4 decoding on Hopper, with fused SwiGLU, requantization, and optional shared experts. On 8 H20 GPUs GC-MoE accelerates DeepSeek-V4.1-Flash decode by 1.188x on average on both the routed and shared-expert suites (peaks 1.296x and 1.353x); on 8 B300 GPUs it accelerates GLM-5.2 and DeepSeek-V4.1-Flash by 1.106x (1.123x with fused shared experts, peak 1.209x), beating Mega MoE at all 54 measured points at matched accuracy. GC-MoE reserves only 20 of 148 SMs on B300 for dispatch.
Image restoration is a fundamental inverse problem that seeks to recover high-quality images from degraded observations. Plug-and-Play (PnP) methods provide a flexible framework for integrating pretrained denoisers into iterative reconstruction, whereas flow matching offers an efficient generative modeling approach based on continuous-time ODE dynamics. Very recently, Martin et al.\ (ICLR, 2025) developed a PnP-Flow method, which integrates the PnP framework with flow matching through a time-dependent denoiser constructed from a pretrained flow matching model.By incorporating a different time schedule and an additional pixel-space projection operator, this paper proposes a novel efficient PnP-Flow algorithm (HP-PnP-Flow) that enjoys rigorous convergence guarantees under mild assumptions, without imposing restrictive conditions on iterates required by the aforementioned work. Numerical experiments on CelebA and AFHQ-Cat demonstrate that the newly proposed HP-PnP-Flow algorithm significantly outperforms PnP-Flow on several restoration tasks such as denoising, deblurring, super-resolution, random inpainting, and box inpainting. Notably, our algorithm achieves comparable or better reconstruction quality with substantially fewer iterations, reducing the iteration count from $500$ to $100$ for deblurring and super-resolution on AFHQ-Cat.
High-fidelity head avatar reconstruction plays a crucial role in AR/VR, gaming, and multimedia content creation. Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated effectiveness in modeling complex geometry with real-time rendering capability and are now widely used in high-fidelity head avatar reconstruction tasks. However, existing 3DGS-based methods still face significant challenges in capturing fine-grained facial expressions and preserving local texture continuity, especially in highly deformable regions. To mitigate these limitations, we propose a novel 3DGS-based framework termed EAvatar for head reconstruction that is both expression-aware and deformation-aware. Our method introduces a sparse expression control mechanism, where a small number of key Gaussians are used to influence the deformation of their neighboring Gaussians, enabling accurate modeling of local deformations and fine-scale texture transitions. Furthermore, we leverage high-quality 3D priors from pretrained generative models to provide a more reliable facial geometry, offering structural guidance that improves convergence stability and shape accuracy during training. Experimental results demonstrate that our method produces more accurate and visually coherent head reconstructions with improved expression controllability and detail fidelity.
Head avatar modeling requires jointly optimizing multiple objectives with different dominant effects on geometry, appearance, and cross-view consistency. However, their relative effectiveness varies across training states, while existing pipelines typically rely on fixed loss weights or handcrafted stage-wise schedules. A central challenge is therefore to identify which optimization direction is more beneficial at each training state. We propose a counterfactual route optimization framework for Gaussian head avatar modeling, which characterizes state-dependent optimization preference from the realized effects of alternative updates rather than predefined heuristic weighting. Starting from the same training state, we perform short-horizon route-restricted lookahead over geometry, appearance, and joint update routes and evaluate their outcomes under a unified utility. The resulting counterfactual evidence is factorized into a geometry--appearance preference and a residual joint advantage, separately capturing the relative preference between individual update directions and the additional benefit of coordinated optimization. We further amortize this offline evidence into a lightweight controller that directly estimates the current optimization preference and applies bounded modulation to the training objectives during full avatar optimization. Experiments on the NeRSemble dataset validate the effectiveness of the proposed design, consistently outperforming existing methods while preserving clearer local facial structures and finer details.
We propose Fresco++, a unified optimization framework for fine-grained and view-consistent head avatar reconstruction. Head avatar optimization is typically driven by per-view image supervision, which can lead to premature fitting of unstable high-frequency details and inconsistent local appearance across viewpoints. Fresco++ addresses these challenges by regulating both the progression of visual detail and the formation of cross-view supervision during optimization. For frequency-aware optimization, a progressive curriculum first stabilizes low-frequency structures and then introduces high-frequency constraints to recover fine facial and hair details without amplifying spurious responses at early stages. For cross-view optimization, we introduce Canonical Group Consensus, which associates local observations through shared canonical surface regions and establishes correspondence across different viewpoints. Geometric and visibility-aware screening removes unreliable observations, while the remaining multi-view evidence is aggregated in feature space to form a consensus target for supervising the current rendering. This design enforces local consistency without relying on a specific image-space parameterization and avoids additional rendering of the auxiliary view. Together, the frequency curriculum and canonical consensus provide stable optimization from coarse structures to fine details while maintaining coherent appearance across viewpoints. Extensive experiments on NeRSemble demonstrate improved reconstruction quality and cross-view consistency, while evaluations across diverse avatar representations further confirm the generality and transferability of Fresco++.
Realizing Embodied AI for Science requires robots to translate experimental goals into reliable physical procedures, reducing hazardous and repetitive human work while generating evidence for scientific discovery. Although language-based planning and learned manipulation provide complementary capabilities, their integration requires robots to adapt subsequent operations to actual execution outcomes. For example, an unsuccessful placement changes what must be done next, even when the experimental goal remains unchanged. Existing benchmarks assess planning, manipulation, and laboratory workflows through task completion and reasoning or execution metrics; however, aggregate performance leaves the connection between outcome-dependent choices and physical execution unresolved. To address this evaluation gap, we introduce Labference, a benchmark for evaluating how observed physical outcomes guide subsequent choices and how those choices translate into robotic laboratory actions. Its evaluation protocol is designed to test outcome-dependent behavioral adaptation through controlled evidence variations, while independently measuring operation selection, physical execution, and task completion. Matched Oracle-assisted continuations from shared checkpoints then assess which forms of support improve subsequent choices and task completion. Initial weighing evaluations reveal target-selection errors in the tested VLA policies despite successful object transfers, whereas GPT-driven control achieves correct selection but encounters execution failures and substantial computation costs. Together, these findings highlight the need to connect task-appropriate reasoning with reliable, timely action. By making this connection measurable and diagnosable, Labference supports targeted improvements toward autonomous experimentation for Embodied AI for Science.
Recent vision-language navigation (VLN) systems increasingly adapt pretrained vision-language models (VLMs) into vision-language-action (VLA) policies that map egocentric observations and language instructions directly to navigation actions. Although semantically capable, such action-centric training does not explicitly model how the agent's visual observations evolve under its predicted motion. Generative world-action models (WAMs) jointly predict future observations and actions, yet existing WAMs for continuous VLN do not condition joint future-view and action generation on geometry-aware representations inferred from the observation history. We present WNM-3D, a generative World Navigation Model with 3D scene conditioning for continuous VLN. A frozen feed-forward geometry encoder extracts cross-view representations from monocular RGB history, and a trainable 3D Scene-to-Token Adapter converts them into a fixed-length scene prefix for the world-action Diffusion Transformer. Through block-causal attention, this prefix remains visible to every future video-action block throughout joint denoising, providing persistent geometric context. We train WNM-3D with supervised world-action learning, DAgger-style adaptation on policy-visited states, and Counterfactual DanceGRPO refinement. Across GN-Bench and VLN-CE, WNM-3D consistently outperforms the backbone-matched WNM-2D counterpart, demonstrating gains from 3D scene conditioning across distinct visual domains. WNM-3D further outperforms strong VLN baselines on GN-Bench and demonstrates closed-loop deployment on a physical robot. Component and stage-wise ablations support the Adapter design and the complementary roles of DAgger-SFT and subsequent reward-guided refinement.
Enforcing prescribed global integral constraints in mesh-free neural PDE solvers is challenging in high-dimensional domains. Existing projection methods for spatial integrals are often tied to fixed grids or uniform quadrature, which can conflict with randomly sampled physics-informed neural networks (PINNs) and scale poorly with dimension. High-order differential operators also increase reverse-mode automatic differentiation memory costs. We propose Stochastic Dimension Implicit Functional Projection (SDIFP), a quadrature-level framework for enforcing prescribed first and second spatial moments. SDIFP replaces tensor-product nodal projection by a global affine correction of the neural-network output, with two scalar coefficients determined from a weighted quadrature rule. Under positive target variance and nonzero empirical raw variance, this correction is the nearest-point projection, in the weighted quadrature norm, onto the empirical two-moment constraint set. Thus, the prescribed moments are exact for the selected quadrature rule, while continuum errors are quadrature errors of the corrected field. For decomposable high-dimensional linear operators, SDIFP combines affine moment correction with stochastic operator-subset sampling. With independent residual and derivative sampling and conditionally unbiased coefficient-gradient estimation, the resulting estimator is unbiased for the specified quadrature-based residual objective; the shared-subset fast mode is biased in general. SDIFP avoids tensor-product quadrature for moment enforcement, separates forward quadrature evaluation from the reverse-mode graph, and retains pointwise inference efficiency once the affine coefficients are fixed or precomputed.
While data curation is widely credited with improving large language model training, a basic question remains unresolved: does curation expand what a model can ultimately learn, or merely reach the same solution faster? In this paper, we formalize this distinction between advancement (improving the attainable performance) and acceleration (reaching the same performance with less data). Through an analytic characterization and real-world LLM experiments, we show that advancement requires a correctable train-target gap. When the baseline training and target objectives coincide, no curation policy improves the endpoint, although curation may still accelerate training. This reframes advancement as a property of the curation task rather than the curation method. Under a common gap-centric view of data mixture and selection, we operationalize this condition as the $Δ$-test, an observable discrepancy between the uncurated baseline's training and target losses that is computable prior to expensive curation. Across extensive experiments, an appreciably nonzero $Δ$ predicts persistent gains, whereas gains near $Δ=0$ tend to diminish as the training budget grows. Beyond individual methods, our findings recenter data curation on the task itself and point to target-set construction as a promising direction for shaping what curation can achieve.
Generative search engines (GSEs), which combine large language models (LLMs) with retrieval-augmented generation (RAG), are reshaping how users access Web information and making source visibility and influence in generated responses new optimization objectives. Existing generative search engine optimization (G-SEO) methods seek to improve content visibility, but many rely on direct access to target queries or GSE feedback elicited by those queries. However, Web content is typically optimized and published before future user queries are known, making G-SEO without access to target queries or their associated feedback an important yet underexplored problem. To address this, we propose RAID, a Role-Augmented Intent-Driven G-SEO method that operates without access to target queries or query-dependent GSE feedback. RAID infers a primary retrieval intent from source content, refines it through 4W multi-role deep reflection on prospective searchers' information needs, and uses both the primary and refined intents to jointly plan and rewrite source content. We further extend GEO-BENCH with diverse topic-related query variants, using both the original queries and their variants exclusively for evaluation to assess generalization to queries unseen during optimization. Across 100 samples and 500 evaluation queries, RAID achieves Relative Improvement Scores (RIS) of 3.00\% for Objective Overall and 1.32\% for Subjective Average, exceeding the best corresponding results among the nine predefined GEO rewriting strategies by 1.31 and 0.74 percentage points, respectively. With an optimization considered effective when $\mathrm{RIS}>0$, its effective optimization rates for Objective Overall and Subjective Average average 58.0\%, exceeding the highest corresponding average among the predefined strategies by 4.5 percentage points.
Zhuohang Li, Yiding Ma, Yu Shang, Lei Jin, Shengyuan Wang, Jianjie Fang, Zhaolu Wang, Xvyuan Liu 等 29 位作者
World models have emerged as a central paradigm for embodied intelligence, enabling agents to predict action-conditioned future and reason about environmental dynamics. However, existing embodied world model benchmarks are still largely confined to vision-only prediction, offline embodied applications, and simulator-based evaluation, making them insufficient for assessing increasingly comprehensive world models. In this work, we introduce WorldArena 2.0, an expanded benchmark that systematically broadens embodied world model evaluation along three dimensions: modality, functionality, and platform. Along the modality dimension, WorldArena 2.0 extends evaluation from vision-only to visuotactile modalities, enabling assessment of multimodal perception and prediction. Along the functionality dimension, it extends beyond policy evaluation and planning to assess world models as interactive RL environments for policy optimization. Along the platform dimension, it moves beyond simulator-only evaluation to a diverse suite of simulated and real-world robotic settings across multiple embodiments. Under a standardized protocol, WorldArena 2.0 comprehensively evaluates perceptual quality, interactive utility, and cross-platform performance, providing a comprehensive testbed for tracking progress toward embodied world models. The project page including codes and demos is available at: https://world-arena.ai/.
Ranking functions are a fundamental technique for proving loop termination, but constructing them manually is challenging and requires substantial expertise. We propose RAINFALL, a neuro-symbolic framework which follows the generate-verify-refine paradigm and tightly integrates LLMs with symbolic verification. RAINFALL formulates ranking function synthesis as a code-completion task by inserting placeholders into the input program and constraining the LLM to fill four predefined slots for each loop. The completed program is then checked by a symbolic verifier. If verification succeeds, a valid ranking function is obtained; otherwise, the verifier's diagnostics are translated into natural-language feedback that guides iterative refinement of the LLM's output. We evaluate RAINFALL on two public benchmark suites comprising 358 terminating C programs. Using MiniMax V2.7, RAINFALL synthesizes verifiable ranking functions for 323 programs (90.22%), outperforming all promising baselines. In particular, it solves 12 more programs than PROTON 2.1, the best-performing baseline and winner of the SV-COMP 2026 Termination category. Furthermore, combining RAINFALL with PROTON 2.1 increases the success rate to 96.65% (346/358 programs). We also conduct ablation study confirming individual contributions of placeholder contracts, few-shot examples, symbolic verification, and diagnostic information feedback. Experiments across LLMs further confirm the effectiveness of RAINFALL. Our results suggest a generic, sound framework for ranking-function synthesis and, beyond termination, LLM-assisted synthesis of formally checkable liveness certificates, both of which represent new directions of software engineering research.
World models are emerging as a fundamental component of policy learning for real-world agents, enabling agents to predict future world states and plan actions within a learned representation of the dynamic environment. Existing approaches, however, often model appearance, metric geometry represented by projected LiDAR depth, and ego motion either partially or independently, overlooking the fact that these modalities are different projections of a shared underlying world state. This separation can lead to information loss and cross-modal inconsistency, ultimately limiting downstream performance. Motivated by this unified view of the world, we propose DriveUniTok, a unified multimodal tokenizer for driving world-action models (WAMs) that encodes these modalities in a single decodable continuous latent. We freeze DriveUniTok and adapt a pretrained diffusion model, DriveUniWAM, to predict future unified latents; the tokenizer then decodes them into future appearance, projected LiDAR depth, and ego trajectories. Experiments on NAVSIM demonstrate that our unified design achieves better future prediction and planning performance at lower downstream computational cost than separate-stream alternatives. With reinforcement learning post-training, DriveUniWAM achieves a PDMS of 91.54, reaching state-of-the-art performance among the evaluated WAM-based planners.
Large language models (LLMs) have demonstrated strong capabilities in complex reasoning tasks, yet their autoregressive generation paradigm makes it computationally expensive to explore diverse reasoning paths. In contrast, diffusion language models (DLMs) adopt a parallel, non-autoregressive generation mechanism that enables the efficient production of diverse candidate outputs. Motivated by this complementarity, we explore a collaborative reasoning framework that combines diffusion-based generation with autoregressive evaluation. Specifically, we leverage DLMs to efficiently generate diverse intermediate reasoning thoughts, and employ LLMs as evaluators to assess and select candidates based on their plausibility and correctness. By decoupling proposal generation from evaluation, our framework exploits the strengths of both models: efficient exploration from diffusion models and causally grounded assessment from autoregressive models, which naturally aligns with the divergent-convergent thinking framework in cognitive psychology. Experiments across various mathematical and logical reasoning benchmarks demonstrate that our framework improves inference efficiency while maintaining competitive or superior reasoning accuracy, laying the groundwork for building efficient reasoning architectures. Our code is open-source at https://anonymous.4open.science/r/Diffuse-Thinking-EC60.
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
Jiasen Gao, Xiaoliang Chen Xihua University, Duoqian Miao, Xu Gu, Xianyong Li Xihua University, Yajun Du
Spurious correlations cause deep learning models to rely on predictive shortcuts that hold in the training data but break under distribution shifts, leading to large performance drops for minority groups. Existing strategies often rely on costly group annotations or employ unstable adversarial training. In this paper, we propose Prototype-guided debiasing using Robust Invariant Feature Transformations (PRIFT), a novel framework that mitigates spurious correlations by manipulating latent space geometry. Specifically, we introduce a prototype-guided modeling approach that leverages natural language prompts to represent confounders, transforming abstract biases into interpretable geometric anchors without auxiliary classifiers. Based on these anchors, we introduce a centered projection operator that adaptively purifies representations by removing confounding deviations specific to instances while preserving essential semantic structure. Furthermore, PRIFT can handle confounding factor information at different levels, ranging from true labels to unsupervised latent inference. Experiments on four text classification benchmarks demonstrate the superiority of our method; notably, PRIFT outperforms state-of-the-art baselines and improves worst-group accuracy by over 20% on the CivilComments dataset compared to standard empirical risk minimization.
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
Xiang Cheng Renmin University of China, Yulan Hu Alibaba Group, Xiangwen Zhang, Lu Xu, Lide Tan, Zheng Pan Alibaba Group, Xin Li Alibaba Group, Yong Liu Renmin University of China
Travel planning is a natural real-world task to test large language models’ (LLMs) planning and tool-use abilities. Although prior work has studied LLM performance on travel planning, existing settings still differ from real-world needs, mainly due to limited domain coverage, insufficient modeling of users’ implicit preferences in multi-turn conversations, and a lack of evaluation of agents’ capability boundaries. To mitigate these gaps, we propose \mbox{\textbf{TravelBench}}, a benchmark for \textit{truly real-world} travel planning. We collect user queries, user preferences, and tools from real scenarios, and construct three subtasks—\textit{Single-Turn}, \textit{Multi-Turn}, and \textit{Unsolvable}—to evaluate agents’ three core capabilities in real settings: (1) solving problems independently, (2) interacting with users to elicit implicit preferences, and (3) recognizing the capability boundaries. To enable stable tool invocation and reproducible evaluation, we cache real tool-call results and build a sandbox environment which integrates ten travel-related tools, enabling agents to combine these tools to solve most practical travel planning problems. We evaluate multiple LLMs on TravelBench and find that even advanced models exhibit imbalanced performance across different capabilities. Our further systematic verification demonstrates the stability of the proposed benchmark. TravelBench provides a practical and reproducible benchmark to advance research on LLM agents for real-world travel planning.
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
Chenhui Mao Alibaba Group, Yuanting Lei, Zhixiang Wei, Ming Liang, Zhixiang Wang Peking University, Jingxuan Xu Beijing Jiaotong University, Dajun Chen, Wei Jiang Ant Group 等 9 位作者
Agentic Test-Time Scaling (TTS) has delivered state-of-the-art (SOTA) performance on complex software engineering tasks such as code generation and bug fixing. However, its practical adoption remains limited due to significant computational overhead, primarily driven by two key challenges: (1) the high cost associated with deploying excessively large ensembles, and (2) the lack of a reliable mechanism for selecting the optimal candidate solution—ultimately constraining the performance gains that can be realized. To address these challenges, we propose Entropy-Guided Stepwise Scaling (EGSS), a novel TTS framework that dynamically balances efficiency and effectiveness through entropy-guided adaptive search and robust test-suite augmentation.Extensive experiments on SWE-Bench-Verified demonstrate that EGSS consistently boosts performance by 5–10% across all evaluated models. Specifically, it increases the resolved ratio of Kimi-K2-Intruct from 63.2% to 72.2%, and GLM-4.6 from 65.8% to 74.6%. Furthermore, when paired with GLM-4.6, EGSS achieves new state-of-the-art among open-source large language models. In addition to these accuracy improvements, EGSS reduces inference-time token usage by over 28% compared to existing TTS methods, achieving simultaneous gains in both effectiveness and computational efficiency.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Song Jin, Juntian Zhang, Xun Zhang, Zeying Tian, Fei Jiang Meituan, Guojun Yin Meituan, Wei Lin, Yong Liu Renmin University of China 等 9 位作者
Recent advancements in Vision-Language Models (VLMs) have revolutionized general visual understanding. However, their application in the food domain remains constrained by benchmarks that rely on coarse-grained categories, single-view imagery, and inaccurate metadata. To bridge this gap, we introduce DiningBench, a hierarchical, multi-view benchmark designed to evaluate VLMs across three levels of cognitive complexity: Fine-Grained Classification, Nutrition Estimation, and Visual Question Answering. Unlike previous datasets, DiningBench comprises 3,021 distinct dishes with an average of 5.27 images per entry, incorporating fine-grained “hard” negatives from identical menus and rigorous, verification-based nutritional data. We conduct an extensive evaluation of 29 state-of-the-art open-source and proprietary models. Our experiments reveal that while current VLMs excel at general reasoning, they struggle significantly with fine-grained visual discrimination and precise nutritional reasoning. Furthermore, we systematically investigate the impact of multi-view inputs and Chain-of-Thought reasoning, identifying five primary failure modes. DiningBench serves as a challenging testbed to drive the next generation of food-centric VLM research. All codes are released in https://github.com/meituan/DiningBench.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Junyao Yang, Chen Qian, Wen Shen Tongji University, Yong Liu Renmin University of China, Jing Shao Shanghai AI Laboratory, Dongrui Liu Shanghai Artificial Intelligence Laboratory
Large Reasoning Models (LRMs) with long chain-of-thought reasoning have recently achieved remarkable success. Yet, equipping domain-specialized models with such reasoning capabilities, referred to as “Reasoning + X”, remains a significant challenge. While model merging offers a promising training-free solution, existing methods often suffer from a destructive performance collapse: existing methods tend to both weaken reasoning depth and compromise domain-specific utility. Interestingly, we identify a counter-intuitive phenomenon underlying this failure: reasoning ability predominantly resides in parameter regions with low gradient sensitivity, contrary to the common assumption that domain capabilities correspond to high-magnitude parameters. Motivated by this insight, we propose ReasonAny, a novel merging framework that resolves the reasoning–domain performance collapse through Contrastive Gradient Identification. Experiments across safety, biomedicine, and finance domains show that ReasonAny effectively synthesizes “Reasoning + X” capabilities, significantly outperforming state-of-the-art baselines while retaining robust reasoning performance.
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
Pengxiang Zhao, Guangyi Liu, Yaozhen Liang, Weiqing He, Zhengxi Lu, WenHao Wang, Yuehao Huang, Yuxiang Chai The Chinese University of Hong Kong 等 14 位作者
Shortcuts such as APIs and deep-links have emerged as efficient complements to flexible GUI operations, fostering a promising hybrid paradigm for MLLM-based mobile automation. However, systematic evaluation of GUI–shortcut hybrid agents remains largely underexplored. To bridge this gap, we introduce MAS-Bench, a benchmark that pioneers the evaluation of GUI-shortcut hybrid agents with a specific focus on the mobile domain. Beyond merely using predefined shortcuts, MAS-Bench assesses an agent’s capability to autonomously generate shortcuts by discovering and creating reusable, low-cost workflows. It features 139 complex tasks across 11 real-world applications, a knowledge base of 88 predefined shortcuts (APIs, deep-links, RPA scripts), and 9 evaluation metrics. Experiments demonstrate that hybrid agents achieve up to 68.3% success rate and 39% greater execution efficiency than GUI-only counterparts. Furthermore, our evaluation framework effectively reveals the quality gap between predefined and agent-generated shortcuts, validating its capability to assess shortcut generation methods. MAS-Bench addresses the lack of systematic benchmarks for GUI-shortcut hybrid mobile agents, providing a foundational platform for future advancements in creating more efficient and robust intelligent agents.
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
Sheng Zhang, Junyi Li, Yingyi Zhang, Pengyue Jia City University of Hong Kong, Yichao Wang, Xiaowei Qian City University of Hong Kong, Wenlin Zhang City University of Hong Kong, Maolin Wang 等 10 位作者
Recent advances in large language models (LLMs) have scaled the potential for reasoning and agentic search, wherein models autonomously plan, retrieve, and reason over external knowledge to answer complex queries. However, the iterative think–search loop accumulates long system memories, leading to memory dilution problem. In addition, existing memory management methods struggle to capture fine-grained semantic relations between queries and documents and often lose substantial information. Therefore, we propose MemSearch-o1, an agentic search framework built on reasoning-aligned memory growth and retracing. MemSearch-o1 dynamically grows fine-grained memory fragments from memory seed tokens from the queries, then retraces and deeply refines the memory via a contribution function, and finally reorganizes a globally connected memory path. This shifts memory management from stream-like concatenation to structured, token-level growth with path-based reasoning. Experiments on eight benchmark datasets show that MemSearch-o1 substantially mitigates memory dilution, and more effectively activates the reasoning potential of diverse LLMs, establishing a solid foundation for memory-aware agentic intelligence.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Haotian Xu, Yue Hu National University of Defense Technology, Zhengqiu Zhu National University of Defense Technology, Chen Gao Tsinghua University, Ziyou Wang, Junreng Rao, Wenhao Lu, Weishi Li 等 10 位作者
Cross-view spatial reasoning is essential for embodied AI, underpinning spatial understanding, mental simulation and planning in complex environments. Existing benchmarks primarily emphasize indoor or street settings, overlooking the unique challenges of open-ended urban spaces characterized by rich semantics, complex geometries, and view variations. To address this, we introduce CityCube, a systematic benchmark designed to probe cross-view reasoning capabilities of current VLMs in urban settings. CityCube integrates four viewpoint dynamics to mimic camera movements and spans a wide spectrum of perspectives from multiple platforms, e.g., vehicles, drones and satellites. For a comprehensive assessment, it features 5,022 meticulously annotated multi-view QA pairs categorized into five cognitive dimensions and three spatial relation expressions. A comprehensive evaluation of 33 VLMs reveals a significant performance disparity with humans: even large-scale models struggle to exceed 54.1% accuracy, remaining 34.2% below human performance. By contrast, small-scale fine-tuned VLMs achieve over 60.0% accuracy, highlighting the necessity of our benchmark. Further analyses indicate the task correlations and fundamental cognitive disparity between VLMs and human-like reasoning.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zhicong Li, Lingjie Jiang Peking University, Yulan Hu Alibaba Group, Xingchen Zeng, Yixia Li, Xiangwen Zhang, Guanhua Chen Southern University of Science and Technology, Zheng Pan Alibaba Group 等 10 位作者
Critique-guided reinforcement learning (RL) has emerged as a powerful paradigm for training LLM agents by augmenting sparse outcome rewards with natural-language feedback. However, current methods often rely on static or offline critic models, which fail to adapt as the policy evolves. In on-policy RL, the agent’s trajectory distribution and error patterns shift over time, causing stationary critics to become stale and providing feedback of diminishing utility. To address this, we introduce ECHO (Evolving Critic for Hindsight-Guided Optimization), a framework that jointly optimizes the policy and critic through a synchronized co-evolutionary loop. ECHO utilizes a cascaded rollout mechanism where the critic generates multiple diagnoses for an initial trajectory, followed by policy refinement to enable group-structured advantage estimation. We address the challenge of learning plateaus via a saturation-aware gain shaping objective, which rewards the critic for inducing incremental improvements in high-performing trajectories. By employing synchronized dual-track GRPO updates, ECHO ensures the critic’s feedback stays synchronized with the evolving policy. Experimental results show that ECHO yields more stable training and higher long-horizon task success across open-world environments.
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
Zihang Liu, Fang Zhouhua, Hui Liu, Zhiwei Liu Alibaba Group, Yong Li Alibaba Group, Haishuai Wang Zhejiang University
Large reasoning models (LRMs) achieve strong performance on complex tasks by generating intermediate reasoning before the final answer, yet they remain prone to reasoning hallucinations such as subtle arithmetic or constraint-violation errors. Prior hallucination detectors often rely on external verification or local token-level signals, which are limited for LRMs and largely overlook whether the cross-phase information flow from reasoning to answering is structurally robust. We propose Routing Focus Score (RFS), a step-level indicator that measures how strongly cross-step attention routing aligns with semantic proximity derived from hidden-state cosine similarity. We further design RFS-Guard, a lightweight hallucination detection framework based on RFS. Empirically, we observe that higher reasoning–answer RFS is consistently associated with higher hallucination risk, suggesting a routing-collapse failure mode where models might prefer self-confirmation loops and suppress the ability to audit their own generations. Experimental results across multiple domains and models demonstrate the superiority of RFS-Guard for detecting and localizing hallucinations in LRMs without requiring external tools or repeated sampling.
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
AI 速览：ToolACE提出一种自动化的智能体流水线，用于生成准确、复杂且多样的工具学习数据，以提升大语言模型的函数调用能力。
Function calling significantly extends the application boundary of large language models, where high-quality and diverse training data is critical for unlocking this capability. However, real function-calling data is quite challenging to collect and annotate, while synthetic data generated by existing pipelines tends to lack coverage and accuracy. In this paper, we present ToolACE, an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data. ToolACE leverages a novel self-evolution synthesis process to curate a comprehensive API pool of 26,507 diverse APIs. Dialogs are further generated through the interplay among multiple agents, guided by a formalized thinking process. To ensure data accuracy, we implement a dual-layer verification system combining rule-based and model-based checks. We demonstrate that models trained on our synthesized data, even with only 8B parameters, achieve state-of-the-art performance on the Berkeley Function-Calling Leaderboard, rivaling the latest GPT-4 models. Our model and a subset of the data are publicly available at https://huggingface.co/Team-ACE.