Searcharxiv⌕ Search

arXiv subjects

An Zhang

Publications and source records attributed to An Zhang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Learning to Act under Noise: Enhancing Agent Robustness via Noisy Environments

Recent advances in large language models (LLMs) have facilitated the widespread deployment of LLMs as interactive agents capable of reasoning, planning, and tool use. Despite strong performance on existing benchmarks, such agents often exhibit notable degradation when deployed in real-world settings, where environments are inherently stochastic and imperfect. We argue that this discrepancy arises from a fundamental mismatch between idealized training settings and real-world interaction dynamics, where current paradigms rely on carefully curated task instructions and stable, well-controlled environments. To address this gap, we propose NoisyAgent, an agentic training framework that explicitly incorporates environmental imperfections into the agent learning process. We identify two major sources of interaction noise in real-world scenarios: user noise, which captures ambiguity and variability in user interaction, and tool noise, which reflects failures and anomalies in tool execution. We introduce such perturbations into the training pipeline by modifying user interaction patterns and simulating tool execution results within the training environment. To stabilize training while encouraging agents to handle increasingly challenging imperfections, noise is applied to only a subset of rollouts and progressively increased in difficulty as the model adapts to the current noise level. Extensive experiments demonstrate that our approach consistently improves agent robustness under noisy and dynamic environments. Our analysis reveals that training under noise conditions also yields performance gains on idealized benchmarks, suggesting that controlled exposure to environmental noise promotes more generalizable reasoning and decision-making behaviors. Our findings highlight the importance of modeling interaction imperfections for bridging the gap between agent training and real-world deployment.

cs.AI↗
arXiv2026

Covering vertices by sequential stars

We study the problem of covering the maximum number of vertices in a graph by a collection of vertex-disjoint stars, each with a number of satellites in a given interval $[k, \ell]$, where $1 \le k < \ell$ and $\ell$ can be infinity. This is referred to as sequential {\sc $[k, \ell]$-Star Packing} problem. It is solvable in polynomial time when $k = 1$, but becomes strongly NP-hard when $k \ge 2$. In this paper, we propose either the first or an improved approximation algorithm for the following four sequential settings: 1) a $\frac {k+1}2$-approximation algorithm when $k \ge 3$ and $\ell = \infty$, improving the previous best ratio of $\frac {(k+1)^2}{2k+1}$; 2) a $\frac 43$-approximation algorithm when $k = 2$ and $\ell = \infty$, improving the previous best ratio of $\frac 32$; 3) the first $(1 + \frac \ell{\ell+1})$-approximation algorithm when $2 = k < \ell$; and 4) the first $(1 + \max\left\{\frac {k-1}2, \frac {(k+1) \ell}{3 (\ell+1)}\right\})$-approximation algorithm when $3 \le k < \ell$. Besides the main algorithmic techniques being local search coupled with amortized analysis, we observe augmenting configurations to bridge two distant neighborhoods for a local improvement operation. Additionally, the problem has been shown APX-hard when $k \ge 3$; we prove its APX-hardness for the last remaining case where $k = 2$.

cs.DS↗
arXiv2026

MemOCR: Layout-Aware Visual Memory for Efficient Long-Horizon Reasoning

Long-horizon agentic reasoning necessitates effectively compressing growing interaction histories into a limited context window. Most existing memory systems serialize history as text, where token-level cost is uniform and scales linearly with length, often spending scarce budget on low-value details. To this end, we introduce MemOCR, a multimodal memory agent that improves long-horizon reasoning under tight context budgets by allocating memory space with adaptive information density through visual layout. Concretely, MemOCR maintains a structured rich-text memory (e.g., headings, highlights) and renders it into an image that the agent consults for memory access, visually prioritizing crucial evidence while aggressively compressing auxiliary details. To ensure robustness across varying memory budgets, we train MemOCR with reinforcement learning under budget-aware objectives that expose the agent to diverse compression levels. Across long-context multi-hop and single-hop question-answering benchmarks, MemOCR outperforms strong text-based baselines and achieves more effective context utilization under extreme budgets.

cs.AI↗
arXiv2026

Denoising Neural Reranker for Recommender Systems

For multi-stage recommenders in industry, a user request would first trigger a simple and efficient retriever module that selects and ranks a list of relevant items, then the recommender calls a slower but more sophisticated reranking model that refines the item list exposure to the user. To consistently optimize the two-stage retrieval reranking framework, most efforts have focused on learning reranker-aware retrievers. In contrast, there has been limited work on how to achieve a retriever-aware reranker. In this work, we provide evidence that the retriever scores from the previous stage are informative signals that have been underexplored. Specifically, we first empirically show that the reranking task under the two-stage framework is naturally a noise reduction problem on the retriever scores, and theoretically show the limitations of naive utilization techniques of the retriever scores. Following this notion, we derive an adversarial framework DNR that associates the denoising reranker with a carefully designed noise generation module. The resulting DNR solution extends the conventional score error minimization loss with three augmented objectives, including: 1) a denoising objective that aims to denoise the noisy retriever scores to align with the user feedback; 2) an adversarial retriever score generation objective that improves the exploration in the retriever score space; and 3) a distribution regularization term that aims to align the distribution of generated noisy retriever scores with the real ones. We conduct extensive experiments on three public datasets and an industrial recommender system, together with analytical support, to validate the effectiveness of the proposed DNR.

cs.IR↗
arXiv2026

GLT-PEFT: Gated Lie-Tucker Parameter-Efficient Fine-Tuning for Alzheimer's Disease Diagnosis with Hippocampal Segmentation Pretraining

Parameter-efficient fine-tuning (PEFT) has emerged as a promising paradigm for adapting pretrained models under limited data conditions. However, most existing PEFT methods are designed for matrix-structured parameters and are not well suited for high-dimensional convolutional kernels in medical imaging models. Moreover, they typically rely on additive updates and lack mechanisms to preserve the geometric structure of pretrained parameters, while multiplicative (geometry-aware) updates are difficult to integrate within a unified framework. To address this issue, this paper proposes GLT-PEFT, a gated Lie-Tucker parameter-efficient fine-tuning framework for Alzheimer's disease (AD) diagnosis. The proposed approach transfers a hippocampal segmentation pretrained model to a downstream classification task. Tucker decomposition enables tensor-aware low-rank adaptation of 3D convolutional kernels, while Lie group-based transformations provide structure-preserving multiplicative updates. A gating mechanism further reconciles additive and multiplicative update forms, resulting in a unified and more stable fine-tuning strategy. Extensive experiments demonstrate that GLT-PEFT achieves effective cross-task transfer while significantly reducing trainable parameters, highlighting its effectiveness for efficient and robust adaptation in medical imaging models.

cs.CV↗
arXiv2026

Reasoning Can Be Restored by Correcting a Few Decision Tokens

Large reasoning models (LRMs) substantially outperform their base LLM counterparts on challenging reasoning benchmarks, yet it remains poorly understood where base models go wrong during token-by-token generation and how to narrow this gap efficiently. We study the base-reasoning gap through quantifying token-level distributional disagreement between a base model and a stronger reasoning model using likelihood-based divergences. Across benchmarks, we find that the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty -- suggesting that base models fail mainly at early planning points that steer the subsequent reasoning trajectory. Building on these findings, we propose disagreement-guided token intervention, a simple inference-time delegation scheme that performs a one-token takeover by the reasoning model only at high-disagreement positions and immediately switches back to the base model. With a small intervention budget, this sparse delegation substantially recovers and can even surpass the performance of a same-size reasoning model on challenging reasoning tasks. Code is available at https://github.com/AlphaLab-USTC/RRTokenIntervention.

cs.AI↗
arXiv2026

Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion

In the realm of multi-objective alignment for large language models, balancing disparate human preferences often manifests as a zero-sum conflict. Specifically, the intrinsic tension between competing goals dictates that aggressively optimizing for one metric (e.g., helpfulness) frequently incurs a substantial penalty on another (e.g., harmlessness). While prior work mainly focuses on data selection, parameter merging, or algorithmic balancing during training, these approaches merely force compromises between divergent preferences along a fixed Pareto frontier, failing to fundamentally resolve the inherent trade-off. In this work, we approach this problem from a novel perspective of multi-dimensional rewards. By scaling up the model's rollouts and analyzing the outputs across different reward dimensions, we arrive at a critical conclusion: the conflict among multiple objectives stems from the fact that the prompt itself inherently restricts the achievable multi-dimensional rewards. Based on this core observation, we propose MORA: Multi-Objective Reward Assimilation. Specifically, MORA isolates single-reward prompts through pre-sampling and expands their reward diversity by rewriting the original questions to incorporate multi-dimensional intents. Extensive experiments demonstrate that: (1) in sequential alignment, MORA achieves single-preference improvements ranging from 5% to 12.4%, with exceptional gains in harmlessness, after multiple-preference alignment across helpful, harmless, and truthful dimensions. (2) In simultaneous alignment, MORA achieves an average overall reward improvement of 4.6%. Our codes are available at https://github.com/Shiying-Huang/MORA-MPA.

cs.AI↗
arXiv2026

Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning

A persistent skill library allows language model agents to reuse successful strategies across tasks. Maintaining such a library requires three coupled capabilities. The agent selects a relevant skill, utilizes it during execution, and distills new skills from experience. Existing methods optimize these capabilities in isolation or with separate reward sources, resulting in partial and conflicting evolution. We propose Skill1, a framework that trains a single policy to co-evolve skill selection, utilization, and distillation toward a shared task-outcome objective. The policy generates a query to search the skill library, re-ranks candidates to select one, solves the task conditioned on it, and distills a new skill from the trajectory. All learning derives from a single task-outcome signal. Its low-frequency trend credits selection and its high-frequency variation credits distillation. Experiments on ALFWorld and WebShop show that Skill1 outperforms prior skill-based and reinforcement learning baselines. Training dynamics confirm the co-evolution of the three capabilities, and ablations show that removing any credit signal degrades the evolution.

cs.AI↗
arXiv2026

Strichartz estimates for orthonormal systems on compact manifolds: the non-sharp region

We establish new Strichartz estimates for orthonormal systems on compact Riemannian manifolds in the non-sharp admissible region of exponents, covering wave, Klein-Gordon, and fractional Schrödinger equations. Our approach combines the result of Wang-Zhang-Zhang \cite{wang2025strichartz} on the sharp admissible line with a Lieb-Sobolev inequality derived from a recent Cwikel estimate due to Sukochev-Yang-Zanin \cite{sukochev2025singular}, along with an alternative globalization method based on localized weak Lorentz estimates. Our results extend the Euclidean results of Bez-Hong-Lee-Nakamura-Sawano \cite{bez2019strichartz} and Bez-Lee-Nakamura \cite{bez2021strichartz}, as well as the classical single-function estimates on manifolds due to Kapitanski \cite{kapitanski1989some}, Burq-Gérard-Tzvetkov \cite{MR2058384}, and Dinh \cite{dinh2016strichartz}.

math.CA↗
arXiv2026

Internalizing Safety Understanding in Large Reasoning Models via Verification

While explicit Chain-of-Thought (CoT) empowers large reasoning models (LRMs), it enables the generation of riskier final answers. Current alignment paradigms primarily rely on externally enforced compliance, optimizing models to detect malicious prompts rather than evaluating the safety of their own outputs. We argue that this approach remains largely behavioral: our empirical analysis reveals that ostensibly aligned models lack intrinsic safety understanding, often failing to verify their own response safety and remaining vulnerable to adversarial jailbreaks. To address this fundamental limitation, we propose Safety Internal (SInternal), a framework that internalizes safety specifications by training LRMs exclusively on safety verification tasks to critique their own generated answers using expert reasoning trajectories. We demonstrate that learning to verify induces a strong generalization for response safety, significantly enhancing robustness against out-of-domain jailbreaks. Furthermore, when combined with reinforcement learning, SInternal serves as a superior initialization compared to standard supervised fine-tuning, suggesting that internalizing safety understanding creates a more robust foundation for alignment than merely mimicking safe behaviors. Our codes are available at https://github.com/AlphaLab-USTC/SInternal

cs.AI↗
arXiv2026

Self-ReSET: Learning to Self-Recover from Unsafe Reasoning Trajectories

Large Reasoning Models possess remarkable capabilities for self-correction in general domain; however, they frequently struggle to recover from unsafe reasoning trajectories under adversarial attacks. Existing alignment methods attempt to mitigate this vulnerability by fine-tuning the model on expert data including reflection traces or adversarial prefixes. Crucially, these approaches are often hindered by static training data which inevitably deviate from model's dynamic, on-policy reasoning traces, resulting in model hardly covering its vast generation space and learning to recover from its own failures. To bridge this gap, we propose Self-ReSET, a pure reinforcement learning framework designed to equip LRMs with the intrinsic capacity to recover from their own safety error trajectories, which are subsequently reused as an initial state for reinforcement learning. Extensive experiments across various LRMs and benchmarks demonstrate that Self-ReSET significantly enhances robustness against adversarial attacks especially out-of-distribution (OOD) jailbreak prompts while maintaining general utility, along with efficient data utilization. Further analysis reveals that our method effectively fosters self-recovery patterns, enabling models to better identify and recover from unsafe intermediate error states back to benign paths. Our codes and data are available at https://github.com/Ing1024/Self-ReSET.

cs.AI↗
arXiv2026

Near-Tight Approximation Algorithms for Bottleneck Multiple Knapsack Problems

In the bottleneck multiple knapsack problem, we are given a set of items and a set of knapsacks, where each item has a profit and a weight, and each knapsack has a capacity. Our goal is to assign items to knapsacks so as to maximize the minimum profit received by any knapsack subject to the capacity constraint. When all knapsacks have identical capacity, we give a $(\frac{2}{3} - \varepsilon)$-approximation algorithm for any constant $\varepsilon > 0$. This result almost matches the $(\frac{2}{3} + \varepsilon)$ inapproximability bound for the bottleneck multiple subset sum problem (Caprara et al., 2000). When the knapsacks can have arbitrary capacities, we propose a $(\frac{1}{2} - \varepsilon)$-approximation algorithm for any constant $\varepsilon > 0$. We also prove a hardness bound of $(\frac{1}{2} + \varepsilon)$ for any constant $\varepsilon > 0$.

cs.DS↗
arXiv2026

Joint Optimization of Multi-agent Memory System

Memory systems are critical for LLMs, mitigating context window limitations and supporting long-horizon user-LLM interactions. Such systems typically comprise multiple agents responsible for memory construction and retrieval. Existing approaches often optimize each agent independently under a shared global objective (e.g., downstream QA accuracy), treating other agents as a static environment. However, this design has two key limitations: (1) independent optimization ignores inter-agent dependencies and lacks agents' co-adaptation, and (2) relying solely on sparse global rewards provides limited guidance for optimizing specialized agents and causes ambiguous credit assignment. These may ultimately limit agents' effective collaboration in the memory system. To address these limitations, we propose CoMAM, a joint optimization framework that promotes collaboration among agents via end-to-end reinforcement learning and an adaptive credit assignment mechanism. Specifically, we model the multi-agent pipeline as a Markov decision process (MDP) to expose inter-agent dependencies during end-to-end training. Agents are then jointly optimized using a combination of their local task reward and an adaptively weighted global reward, enabling agents to co-adapt while receiving targeted feedback for their respective roles. Experiments show that CoMAM consistently outperforms leading memory systems, validating the effectiveness of the joint optimization framework.

cs.MA↗
arXiv2026

SafeSieve: From Heuristics to Experience in Progressive Pruning for LLM-based Multi-Agent Communication

LLM-based multi-agent systems exhibit strong collaborative capabilities but often suffer from redundant communication and excessive token overhead. Existing methods typically enhance efficiency through pretrained GNNs or greedy algorithms, but often isolate pre- and post-task optimization, lacking a unified strategy. To this end, we present SafeSieve, a progressive and adaptive multi-agent pruning algorithm that dynamically refines the inter-agent communication through a novel dual-mechanism. SafeSieve integrates initial LLM-based semantic evaluation with accumulated performance feedback, enabling a smooth transition from heuristic initialization to experience-driven refinement. Unlike existing greedy Top-k pruning methods, SafeSieve employs 0-extension clustering to preserve structurally coherent agent groups while eliminating ineffective links. Experiments across benchmarks (SVAMP, HumanEval, etc.) showcase that SafeSieve achieves 94.01% average accuracy while reducing token usage by 12.4%-27.8%. Results further demonstrate robustness under prompt injection attacks (1.23% average accuracy drop). In heterogeneous settings, SafeSieve reduces deployment costs by 13.3% while maintaining performance. These results establish SafeSieve as an efficient, GPU-free, and scalable framework for practical multi-agent systems. Our code can be found here: https://github.com/csgen/SafeSieve

cs.MA↗
arXiv2026

Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agents

Large language models face challenges in long-context question answering, where key evidence of a query may be dispersed across millions of tokens. Existing works equip large language models with a memory buffer that is dynamically updated via a linear document scan, also known as the "memorize while reading" methods. While this approach scales efficiently, it suffers from pruning of latent evidence, information loss through overwriting, and sparse reinforcement learning signals. To tackle these challenges, we present ReMemR1, which integrates the mechanism of memory retrieval into the memory update process, enabling the agent to selectively callback historical memories for non-linear reasoning. To further strengthen training, we propose a multi-level reward design, which combines final-answer rewards with dense, step-level signals that guide effective memory use. Together, these contributions mitigate information degradation, improve supervision, and support complex multi-hop reasoning. Extensive experiments demonstrate that ReMemR1 significantly outperforms state-of-the-art baselines on long-context question answering while incurring negligible computational overhead, validating its ability to trade marginal cost for robust long-context reasoning.

cs.CL↗
arXiv2026

Quantile Advantage Estimation: Stabilizing RLVR for LLM Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) strengthens LLM reasoning, but training often oscillates between {entropy collapse} and {entropy explosion}. We trace both hazards to the mean baseline used in value-free RL (e.g., GRPO and DAPO), which improperly penalizes negative-advantage samples under reward outliers. We propose {Quantile Advantage Estimation} (QAE), replacing the mean with a group-wise K-quantile baseline. QAE induces a response-level, two-regime gate: on hard queries (p <= 1 - K) it reinforces rare successes, while on easy queries (p > 1 - K) it targets remaining failures. Under first-order softmax updates, we prove {two-sided entropy safety}, giving lower and upper bounds on one-step entropy change that curb explosion and prevent collapse. Empirically, this minimal modification stabilizes entropy, sparsifies credit assignment (with tuned K, roughly 80% of responses receive zero advantage), and yields sustained pass@1 gains on Qwen3-8B/14B-Base across AIME 2024/2025 and AMC 2023. These results identify {baseline design} -- rather than token-level heuristics -- as the primary mechanism for scaling RLVR.

cs.LG↗
arXiv2026

Transport and Merge: Cross-Architecture Merging for Large Language Models

Large language models (LLMs) achieve strong capabilities by scaling model capacity and training data, yet many real-world deployments rely on smaller models trained or adapted from low-resource data. This gap motivates the need for mechanisms to transfer knowledge from large, high-resource models to smaller, low-resource targets. While model merging provides an effective transfer mechanism, most existing approaches assume architecture-compatible models and therefore cannot directly transfer knowledge from large high-resource LLMs to heterogeneous low-resource targets. In this work, we propose a cross-architecture merging framework based on optimal transport (OT) that aligns activations to infer cross-neuron correspondences between heterogeneous models. The resulting transport plans are then used to guide direct weight-space fusion, enabling effective high-resource to low-resource transfer using only a small set of inputs. Extensive experiments across low-resource languages and specialized domains demonstrate consistent improvements over target models.

cs.CL↗
arXiv2026

Do LLMs and VLMs Share Neurons for Inference? Evidence and Mechanisms of Cross-Modal Transfer

Large vision-language models (LVLMs) have rapidly advanced across various domains, yet they still lag behind strong text-only large language models (LLMs) on tasks that require multi-step inference and compositional decision-making. Motivated by their shared transformer architectures, we investigate whether the two model families rely on common internal computation for such inference. At the neuron level, we uncover a surprisingly large overlap: more than half of the top-activated units during multi-step inference are shared between representative LLMs and LVLMs, revealing a modality-invariant inference subspace. Through causal probing via activation amplification, we further show that these shared neurons encode consistent and interpretable concept-level effects, demonstrating their functional contribution to inference. Building on this insight, we propose Shared Neuron Low-Rank Fusion (SNRF), a parameter-efficient framework that transfers mature inference circuitry from LLMs to LVLMs. SNRF profiles cross-model activations to identify shared neurons, computes a low-rank approximation of inter-model weight differences, and injects these updates selectively within the shared-neuron subspace. This mechanism strengthens multimodal inference performance with minimal parameter changes and requires no large-scale multimodal fine-tuning. Across diverse mathematics and perception benchmarks, SNRF consistently enhances LVLM inference performance while preserving perceptual capabilities. Our results demonstrate that shared neurons form an interpretable bridge between LLMs and LVLMs, enabling low-cost transfer of inference ability into multimodal models. Our code is available at [https://github.com/chenhangcuisg-code/Do-LLMs-VLMs-Share-Neurons](https://github.com/chenhangcuisg-code/Do-LLMs-VLMs-Share-Neurons).

cs.CL↗