Before AI can be trusted to discover new science, it should be able to reproduce what science already knows. Reproduction provides a rigorous checkpoint for an AI scientist: the target claim is known, yet recovering it still requires translating an incomplete scientific description into an executable experiment and producing evidence that can be independently verified. Existing research-reproduction benchmarks largely focus on AI and machine learning, where experiments share relatively homogeneous software stacks. Extending this capability across scientific disciplines introduces a different challenge: agents must operate domain-specific scientific software, make physical and numerical choices, diagnose long-running computations, and distinguish valid evidence from merely plausible outputs. We introduce PAPERBENCHX, a benchmark of 93 expert-constructed paper-reproduction tasks spanning 12 research areas across electromagnetics, photonics, chemistry, materials science, biology, and robotics. Each task provides the source paper, a public task specification, and a domain-native scientific environment involving specialized scientific software. Agents must reconstruct a key experiment, execute and diagnose it, and deliver a runnable workflow that supports the target scientific claim. To evaluate scientific validity rather than surface-level agreement, we replay every submission from scratch and score only regenerated evidence against expert-validated criteria. Across 10 frontier agent configurations, the best configuration fully reproduces only 14% tasks under the 90% threshold. Meanwhile, mean rubric scores range from 27.53% to 59.08%, indicating that substantial partial progress rarely translates into complete scientific reproduction. These results reveal a large gap between paper understanding and reliably producing reproducible scientific evidence.
We study satisficing regret minimization in bandits with knapsacks (BwK), where a learner sequentially selects randomized policies, i.e., distributions over arms, to obtain mean rewards above a satisficing threshold as frequently as possible while respecting finite resource budgets. The performance is measured by satisficing regret, the cumulative deficits of the selected policies' mean rewards below this threshold. Resource constraints introduce two difficulties: an algorithm can incur linear satisficing regret even with small standard regret as deficits to the threshold cannot be compensated, and random consumption can cause early resource depletion even when policies are feasible in expectation. To address these challenges, we propose SELECTwK, a general algorithmic template that uses a BwK learning oracle algorithm with sub-linear expected (standard) regret upper bound to search for and test candidate satisficing policies (i.e., policies whose mean rewards exceed satisficing threshold and whose expected resource consumption is feasible over the time horizon). Our regret analysis, supported by an LP-duality argument, establishes that SELECTwK attains constant expected satisficing regret when a satisficing policy exists. We further introduce a stronger regret condition that bounds the expected sum of positive reward deficits to the optimal one at each time step, under which SELECTwK achieves a sharper constant bound together with a sub-linear bound in the time horizon. As a complement, SELECTwK still preserves the oracle's sub-linear expected standard regret bound if the satisficing threshold is unattainable. Finally, we establish the oracle conditions and corresponding satisficing regret bound for existing finite-armed BwK and dynamic pricing algorithms, and conduct numerical experiments on dynamic pricing with limited inventory to demonstrate the empirical performance of SELECTwK.
We study in-context learning problems where a Transformer is pretrained on tasks drawn from a mixture distribution $π=\sum_{α\in\mathcal{A}} λ_α π_α$, called the pretraining prior, in which each mixture component $π_α$ is a distribution on tasks of a specific difficulty level indexed by $α$. Our goal is to understand the performance of the pretrained Transformer when evaluated on a different test distribution $μ$, consisting of tasks of fixed difficulty $β\in\mathcal{A}$, and with potential distribution shift relative to $π_β$, subject to the chi-squared divergence $χ^2(μ,π_β)$ being at most $κ$. In particular, we consider nonparametric regression problems with random smoothness, and multi-index models with both random smoothness and random effective dimension. We prove that a large Transformer pretrained on sufficient data achieves the optimal rate of convergence corresponding to the difficulty level $β$, uniformly over test distributions $μ$ in the chi-squared divergence ball. Thus, the pretrained Transformer is able to achieve faster rates of convergence on easier tasks and is robust to distribution shift at test time. Finally, we prove that even if an estimator had access to the test distribution $μ$, the convergence rate of its expected risk over $μ$ could not be faster than that of our pretrained Transformers, thereby providing a more appropriate optimality guarantee than minimax lower bounds.
Zihan Gu, Lulu Zhao, Tianyi Ma, Hua Zhang, Wenbo Su, Yanan Cao, Yue Hu
Grokking is a delayed transition from memorisation to generalisation. A network first fits observed examples while remaining near chance on held-out inputs, then generalises abruptly. Prior work has focused mainly on this delay, finding that the network's main loss becomes weight decay after the fitting phase. We name the solution to this optimization problem as the minimum-norm endpoint, and study which tasks make all minimum-norm endpoints generalise from partial observations. Specifically, we model pairwise input tasks as edge colourings of $K_{q,q}$, with Latin colourings as our structured family, and classify each observed instance as all-good, mixed, or no-good by its endpoint set. Furthermore, we prove two contrasting results: (1) Finite-abelian-group families are asymptotically all-good from partial observations; (2) Under a uniform random Latin square, all-good tasks become asymptotically rare for observation-based learners, leaving mixed or no-good behaviour. Moreover, experiments show that relabellings of addition generalise despite their dense polynomial representations. Finally, on tasks of composite order, finite-training results suggest that predictors learn a coarse part of the output before the full label.
Yifan Li, Lichi Li, Anh Dao, Xinyu Zhou, Wenjun Huang, Tianyi Ma, Yicheng Qiao, Zheda Mai 等 18 位作者
Visual large language models (VLLMs) are increasingly used as embodied agents, yet industrial warehouses demand more than reaching a goal: agents must navigate safely around moving workers and vehicles. To evaluate embodied agents' goal reaching and safety, we introduce IndustryNav, a simulation environment with 24 warehouse scenes for closed-loop navigation to specified destinations. Alongside task success and efficiency, we introduce safety-aware metrics that capture collisions and near-collision warnings along each trajectory. Experiments with eleven leading open- and closed-source VLLMs and three navigation baselines show that agents can reach their goals yet collide with moving obstacles, exposing a gap between success and safe navigation. This gap matters in safety-critical settings, where a single collision can end the task. IndustryNav also supports policy learning through automated oracle trajectories and simulator interaction, with held-out scenes for generalization tests. These experiments show that policies can succeed on new routes in familiar warehouses yet fail to transfer to held-out layouts, making cross-scene generalization an open challenge.
Han Zhang, Zihan Gu Institute of Information Engineering, Chinese Academy of Sciences, Zhiyuan Wang, Tianyi Ma, Jiacheng Lu Shanghai Jiao Tong University, Xinyan Zhang independent, Yuhao Wei, Cheng Hua Shanghai Jiaotong University
While Large Language Models (LLMs) achieve high accuracy on established Classical Chinese Poetry benchmarks, it remains challenging to distinguish transferable Linguistic-Aesthetic Reasoning from reliance on familiar pre-training patterns. To address this issue, we introduce Neo-Classic, an evaluation benchmark that combines a constructionist Out-of-Sample (OOS) dataset with a suite of reverse understanding probes. Unlike traditional benchmarks that rely on verification or generation over historical corpora, Neo-Classic comprises strictly metrical poetry authored by contemporary experts, reducing the possibility of direct retrieval. We evaluate state-of-the-art models, including Qwen3-Max, Gemini-3-Pro, and DeepSeek-V3.2, across five behavioral probes designed to test hierarchical constraint satisfaction. Our results reveal two primary limitations. First, a performance gap of 20%–50% emerges when models transition from historical to contemporary texts. Second, models exhibit substantial difficulties in discourse-level ordering tasks, with standard accuracy remaining low (0–13%). Although expert-level guidance improves the performance of reasoning-enhanced models to 36%, a notable gap with human experts persists. These findings suggest that while current LLMs capture local formal patterns, they struggle with global hierarchical planning required for robust Linguistic-Aesthetic Reasoning.
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
Zihan Gu Institute of Information Engineering, Chinese Academy of Sciences, TianYi Zhang University of the Chinese Academy of Sciences, Xinyan Zhang independent, Zhiyuan Wang, Han Zhang, Yuhao Wei, Jiacheng Lu Shanghai Jiao Tong University, Tianyi Ma 等 11 位作者
Model editing provides a promising mechanism for updating large language models (LLMs) without expensive retraining. Existing approaches, particularly locate-and-edit methods based on least-squares optimization, aim to introduce targeted knowledge changes while preserving pre-trained behavior. In this work, we show that this objective is fundamentally fragile under standard single-edit evaluation protocols. We first develop a unified theoretical framework that characterizes activation-based editing as a constrained intervention on intermediate representations. Within this framework, we demonstrate that least-squares edits cannot, in general, isolate target updates from unrelated activations, giving rise to unavoidable interference that accumulates with successive edits. Crucially, this degradation can remain undetected in single-edit settings when assessed using conventional success and locality metrics. To expose such hidden instabilities, we introduce an uncertainty-based evaluation protocol that combines structured semantic perturbations with uncertainty quantification based on Sampling with Perturbation for UQ. By measuring edit-induced growth in aleatoric and epistemic uncertainty, our method reveals local knowledge conflicts that are invisible to existing benchmarks. Extensive experiments across multiple models, datasets, and editing algorithms show that both least-squares and other parameter-update-based methods consistently increase post-edit uncertainty. Together, our results suggest that current evaluation practices substantially overestimate the reliability of single-edit model editing, and that uncertainty-based diagnostics are necessary for assessing edit stability.
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
Zheyuan Zhang, Kaiwen Shi, Zhengqing Yuan, Zehong Wang, Tianyi Ma University of Notre Dame, Keerthiram Murugesan International Business Machines, Vincent Galassi, Chuxu Zhang University of Connecticut 等 9 位作者
Large language models (LLMs) and agent-based frameworks have advanced rapidly, enabling diverse applications. Yet, with the proliferation of models and agentic strategies, practitioners face substantial uncertainty in selecting the best configuration for a downstream task. Prior studies show that different agents and backbones exhibit complementary strengths, and that larger models are not always superior, underscoring the need for adaptive routing mechanisms. Existing approaches to agent routing, however, often emphasize cost efficiency while overlooking the fine-grained contextual and relational structure inherent in QA tasks. In this paper, we propose AgentRouter, a framework that formulates multi-agent QA as a knowledge-graph–guided routing problem supervised by empirical performance signals. Specifically, we convert QA instance into a heterogeneous knowledge graph that jointly encodes queries, contextual entities, and agents, and then train a heterogeneous graph neural network (GNN) to propagate information across node types and produce task-aware routing distributions over agents. By leveraging soft supervision and weighted aggregation of agent outputs, AgentRouter learns principled collaboration schemes that capture the complementary strengths of diverse agents. Extensive experiments demonstrate that our framework consistently outperforms single-agent and ensemble baselines, while generalizing across benchmarks and LLM backbones. These results highlight the effectiveness and robustness of graph-supervised multi-agent routing for question answering. Our code repo is available at https://anonymous.4open.science/r/AgentRouter.
尚无 AI 解读，展示原文摘要
cs.CL · cs.AI · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Tianyi Ma Michigan State University, Yue Zhang, Zehao Wang KU Leuven, Parisa Kordjamshidi Michigan State University
Vision-and-Language Navigation (VLN) poses significant challenges for agents to interpret natural language instructions and navigate complex 3D environments. While recent progress has been driven by large-scale pre-training and data augmentation, current methods still struggle to generalize to unseen scenarios, particularly when complex spatial and temporal reasoning is required. In this work, we propose SkillNav, a modular framework that introduces structured, skill-based reasoning into Transformer-based VLN agents. Our method decomposes navigation into a set of interpretable atomic skills (e.g., Vertical Movement, Area and Region Identification, Stop and Pause), each handled by a specialized agent. To support targeted skill training without manual data annotation, we construct a synthetic dataset pipeline that generates diverse, linguistically natural, skill-specific instruction-trajectory pairs. We then introduce a novel training-free Vision-Language Model (VLM)-based router, which dynamically selects the most suitable agent at each time step by aligning sub-goals with visual observations and previous actions. SkillNav obtains competitive results on commonly used benchmarks and establishes state-of-the-art generalization to the GSA-R2R, a benchmark with novel instruction styles and unseen environments.
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