arXiv subjects

Jiazheng Zhang

Publications and source records attributed to Jiazheng Zhang.

33 records · Page 2Linked to original sources

arXiv2026

EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training

Reinforcement learning (RL) for LLM post-training faces a fundamental design choice: whether to use a learned critic as a baseline for policy optimization. Classical theory favors critic-based methods such as PPO for variance reduction, yet critic-free alternatives like GRPO have gained widespread adoption due to their simplicity and competitive performance. We show that in sparse-reward settings, a learned critic can inject estimation noise that exceeds the state signal it captures, increasing rather than reducing advantage variance. By casting baseline selection as a Kalman filtering problem, we unify PPO and GRPO as two extremes of the Kalman gain and prove that explained variance (EV), computable from a single training batch, identifies the exact boundary: positive EV indicates the critic reduces variance, while zero or negative EV signals that it inflates variance. Building on this insight, we propose Explained Variance Policy Optimization (EVPO), which monitors batch-level EV at each training step and adaptively switches between critic-based and batch-mean advantage estimation, provably achieving no greater variance than the better of the two at every step. Across four tasks spanning classical control, agentic interaction, and mathematical reasoning, EVPO consistently outperforms both PPO and GRPO regardless of which fixed baseline is stronger on a given task. Further analysis confirms that the adaptive gating tracks critic maturation over training and that the theoretically derived zero threshold is empirically optimal.

cs.LG↗

arXiv2026

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

Verifiers have been demonstrated to enhance LLM reasoning via test-time scaling (TTS). Yet, they face significant challenges in complex domains. Error propagation from incorrect intermediate reasoning can lead to false positives for seemingly plausible solutions, while lacking external grounding makes verifiers unreliable on computation or knowledge-intensive tasks. To address these challenges, we propose Agentic Verifier, a framework that transforms reward modeling into a multi-turn, tool-augmented deliberative process. We introduce complementary forward and backward agents: one traces solutions from premises to conclusions, while the other re-checks conclusions against their underlying premises. This bidirectional process enables a comprehensive, reliable, and interpretable assessment of solutions. To facilitate practical deployment, we propose AgentV-RL. Through proactive exploration and reinforcement learning, the verifier autonomously interleaves tool-use with internal reasoning. Extensive experiments show that Agentic Verifier yields consistent performance gains under both parallel and sequential TTS. Notably, our 4B variant surpasses state-of-the-art ORMs by 25.2%, positioning it as a promising paradigm for agentic reward modeling.

cs.CL↗

arXiv2026

Can RL Improve Generalization of LLM Agents? An Empirical Study

Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback. However, most existing evaluations remain largely in-domain: training and testing are conducted in the same environment or even on the same tasks. In real-world deployment, agents may operate in unseen environments with different background knowledge, observation spaces, and action interfaces. To characterize the generalization profile of RFT under such shifts, we conduct a systematic study along three axes: (1) within-environment generalization across task difficulty, (2) cross-environment transfer to unseen environments, and (3) sequential multi-environment training to quantify transfer and forgetting. Our results show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments, which correlates with shifts in both semantic priors and observation/action interfaces. In contrast, sequential training yields promising downstream gains with minimal upstream forgetting, and mixture training across environments improves the overall balance. We further provide detailed analyses and deeper insights, and hope our work helps the community develop and deploy generalizable LLM agents.

cs.AI↗

arXiv2026

Understanding and Mitigating Errors of LLM-Generated RTL Code

Despite limited success in large language model (LLM)-based register-transfer-level (RTL) code generation, the root causes of errors remain poorly understood. To address this, we conduct a comprehensive error analysis, finding that most failures arise not from deficient reasoning, but from a lack of RTL programming knowledge, insufficient circuit understanding, ambiguous specifications, or misinterpreted multimodal inputs. Leveraging in-context learning, we propose targeted correction techniques: a retrieval-augmented generation (RAG) knowledge base to supply domain expertise; design description rules with rule-checking to clarify inputs; external tools to convert multimodal data into LLM-compatible formats; and an iterative simulation-debugging loop for remaining errors. Integrating these into an LLM-based framework yields significant improvement, achieving 98.1% accuracy on the VerilogEval benchmark with DeepSeek-v3.2-Speciale, demonstrating the effectiveness of our approach.

cs.AR↗

arXiv2025

Compression Hacking: A Supplementary Perspective on Informatics Properties of Language Models from Geometric Distortion

Recently, the concept of ``compression as intelligence'' has provided a novel informatics metric perspective for language models (LMs), emphasizing that highly structured representations signify the intelligence level of LMs. However, from a geometric standpoint, the word representation space of highly compressed LMs tends to degenerate into a highly anisotropic state, which hinders the LM's ability to comprehend instructions and directly impacts its performance. We found this compression-anisotropy synchronicity is essentially the ``Compression Hacking'' in LM representations, where noise-dominated directions tend to create the illusion of high compression rates by sacrificing spatial uniformity. Based on this, we propose three refined compression metrics by incorporating geometric distortion analysis and integrate them into a self-evaluation pipeline. The refined metrics exhibit strong alignment with the LM's comprehensive capabilities, achieving Spearman correlation coefficients above 0.9, significantly outperforming both the original compression and other internal structure-based metrics. This confirms that compression hacking substantially enhances the informatics interpretation of LMs by incorporating geometric distortion of representations.

cs.CL↗

arXiv2025

FRABench and UFEval: Unified Fine-grained Evaluation with Task and Aspect Generalization

Evaluating open-ended outputs of Multimodal Large Language Models has become a bottleneck as model capabilities, task diversity, and modality rapidly expand. Existing ``MLLM-as-a-Judge'' evaluators, though promising, remain constrained to specific tasks and aspects. In this paper, we argue that, on one hand, based on the interconnected nature of aspects, learning specific aspects can generalize to unseen aspects; on the other hand, jointly learning to assess multiple visual aspects and tasks may foster a synergistic effect. To this end, we propose UFEval, the first unified fine-grained evaluator with task and aspect generalization for four evaluation tasks -- Natural Language Generation, Image Understanding, Image Generation, and Interleaved Text-and-Image Generation. However, training such a unified evaluator is hindered by the lack of a large-scale, multi-modal, and aspect-level resource. To address this gap, we introduce FRABench, a comprehensive fine-grained evaluation dataset. Specifically, (1) We first construct a hierarchical aspect taxonomy encompassing 112 distinct aspects across the aforementioned four tasks. (2) Based on this taxonomy, we create FRABench, comprising 60.4k pairwise samples with 325k evaluation labels obtained from a combination of human and GPT-4o annotations. (3) Finally, leveraging FRABench, we develop UFEval, a unified fine-grained evaluator. Experiments show that learning on specific aspects enables UFEval to generalize to unseen aspects, and joint learning to assess diverse visual tasks and aspects can lead to substantial mutual benefits.

cs.AI↗

arXiv2025

AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning

Developing autonomous LLM agents capable of making a series of intelligent decisions to solve complex, real-world tasks is a fast-evolving frontier. Like human cognitive development, agents are expected to acquire knowledge and skills through exploration and interaction with the environment. Despite advances, the community still lacks a unified, interactive reinforcement learning (RL) framework that can effectively train such agents from scratch -- without relying on supervised fine-tuning (SFT) -- across diverse and realistic environments. To bridge this gap, we introduce AgentGym-RL, a new framework to train LLM agents for multi-turn interactive decision-making through RL. The framework features a modular and decoupled architecture, ensuring high flexibility and extensibility. It encompasses a wide variety of real-world scenarios, and supports mainstream RL algorithms. Furthermore, we propose ScalingInter-RL, a training approach designed for exploration-exploitation balance and stable RL optimization. In early stages, it emphasizes exploitation by restricting the number of interactions, and gradually shifts towards exploration with larger horizons to encourage diverse problem-solving strategies. In this way, the agent develops more diverse behaviors and is less prone to collapse under long horizons. We perform extensive experiments to validate the stability and effectiveness of both the AgentGym-RL framework and the ScalingInter-RL approach. Our agents match or surpass commercial models on 27 tasks across diverse environments. We offer key insights and will open-source the complete AgentGym-RL framework -- including code and datasets -- to empower the research community in developing the next generation of intelligent agents.

cs.LG↗

arXiv2025

DocFusion: A Unified Framework for Document Parsing Tasks

Document parsing is essential for analyzing complex document structures and extracting fine-grained information, supporting numerous downstream applications. However, existing methods often require integrating multiple independent models to handle various parsing tasks, leading to high complexity and maintenance overhead. To address this, we propose DocFusion, a lightweight generative model with only 0.28B parameters. It unifies task representations and achieves collaborative training through an improved objective function. Experiments reveal and leverage the mutually beneficial interaction among recognition tasks, and integrating recognition data significantly enhances detection performance. The final results demonstrate that DocFusion achieves state-of-the-art (SOTA) performance across four key tasks.

cs.CL↗

arXiv2025

Two Minds Better Than One: Collaborative Reward Modeling for LLM Alignment

Reward models (RMs) play a pivotal role in aligning large language models (LLMs) with human values. However, noisy preferences in human feedback can lead to reward misgeneralization - a phenomenon where reward models learn spurious correlations or overfit to noisy preferences, which poses important challenges to the generalization of RMs. This paper systematically analyzes the characteristics of preference pairs and aims to identify how noisy preferences differ from human-aligned preferences in reward modeling. Our analysis reveals that noisy preferences are difficult for RMs to fit, as they cause sharp training fluctuations and irregular gradient updates. These distinctive dynamics suggest the feasibility of identifying and excluding such noisy preferences. Empirical studies demonstrate that policy LLM optimized with a reward model trained on the full preference dataset, which includes substantial noise, performs worse than the one trained on a subset of exclusively high quality preferences. To address this challenge, we propose an online Collaborative Reward Modeling (CRM) framework to achieve robust preference learning through peer review and curriculum learning. In particular, CRM maintains two RMs that collaboratively filter potential noisy preferences by peer-reviewing each other's data selections. Curriculum learning synchronizes the capabilities of two models, mitigating excessive disparities to promote the utility of peer review. Extensive experiments demonstrate that CRM significantly enhances RM generalization, with up to 9.94 points improvement on RewardBench under an extreme 40\% noise. Moreover, CRM can seamlessly extend to implicit-reward alignment methods, offering a robust and versatile alignment strategy.

cs.LG↗

arXiv2025

Better Process Supervision with Bi-directional Rewarding Signals

Process supervision, i.e., evaluating each step, is critical for complex large language model (LLM) reasoning and test-time searching with increased inference compute. Existing approaches, represented by process reward models (PRMs), primarily focus on rewarding signals up to the current step, exhibiting a one-directional nature and lacking a mechanism to model the distance to the final target. To address this problem, we draw inspiration from the A* algorithm, which states that an effective supervisory signal should simultaneously consider the incurred cost and the estimated cost for reaching the target. Building on this key insight, we introduce BiRM, a novel process supervision model that not only evaluates the correctness of previous steps but also models the probability of future success. We conduct extensive experiments on mathematical reasoning tasks and demonstrate that BiRM provides more precise evaluations of LLM reasoning steps, achieving an improvement of 3.1% on Gaokao2023 over PRM under the Best-of-N sampling method. Besides, in search-based strategies, BiRM provides more comprehensive guidance and outperforms ORM by 5.0% and PRM by 3.8% respectively on MATH-500.

cs.CL↗

arXiv2024

Multi-Programming Language Sandbox for LLMs

We introduce MPLSandbox, an out-of-the-box multi-programming language sandbox designed to provide unified and comprehensive feedback from compiler and analysis tools for Large Language Models (LLMs). It can automatically identify the programming language of the code, compiling and executing it within an isolated sub-sandbox to ensure safety and stability. In addition, MPLSandbox also integrates both traditional and LLM-based code analysis tools, providing a comprehensive analysis of generated code. MPLSandbox can be effortlessly integrated into the training and deployment of LLMs to improve the quality and correctness of their generated code. It also helps researchers streamline their workflows for various LLM-based code-related tasks, reducing the development cost. To validate the effectiveness of MPLSandbox, we integrate it into training and deployment approaches, and also employ it to optimize workflows for a wide range of real-world code-related tasks. Our goal is to enhance researcher productivity on LLM-based code-related tasks by simplifying and automating workflows through delegation to MPLSandbox.

cs.SE↗

arXiv2024

Lightweight Spatial Modeling for Combinatorial Information Extraction From Documents

Documents that consist of diverse templates and exhibit complex spatial structures pose a challenge for document entity classification. We propose KNN-former, which incorporates a new kind of spatial bias in attention calculation based on the K-nearest-neighbor (KNN) graph of document entities. We limit entities' attention only to their local radius defined by the KNN graph. We also use combinatorial matching to address the one-to-one mapping property that exists in many documents, where one field has only one corresponding entity. Moreover, our method is highly parameter-efficient compared to existing approaches in terms of the number of trainable parameters. Despite this, experiments across various datasets show our method outperforms baselines in most entity types. Many real-world documents exhibit combinatorial properties which can be leveraged as inductive biases to improve extraction accuracy, but existing datasets do not cover these documents. To facilitate future research into these types of documents, we release a new ID document dataset that covers diverse templates and languages. We also release enhanced annotations for an existing dataset.

cs.CL↗

arXiv2023

Encoding Node Diffusion Competence and Role Significance for Network Dismantling

Percolation theory shows that removing a small fraction of critical nodes can lead to the disintegration of a large network into many disconnected tiny subnetworks. The network dismantling task focuses on how to efficiently select the least such critical nodes. Most existing approaches focus on measuring nodes' importance from either functional or topological viewpoint. Different from theirs, we argue that nodes' importance can be measured from both of the two complementary aspects: The functional importance can be based on the nodes' competence in relaying network information; While the topological importance can be measured from nodes' regional structural patterns. In this paper, we propose an unsupervised learning framework for network dismantling, called DCRS, which encodes and fuses both node diffusion competence and role significance. Specifically, we propose a graph diffusion neural network which emulates information diffusion for competence encoding; We divide nodes with similar egonet structural patterns into a few roles, and construct a role graph on which to encode node role significance. The DCRS converts and fuses the two encodings to output a final ranking score for selecting critical nodes. Experiments on both real-world networks and synthetic networks demonstrate that our scheme significantly outperforms the state-of-the-art competitors for its mostly requiring much fewer nodes to dismantle a network.

cs.SI↗

arXiv2022

Dismantling Complex Networks by a Neural Model Trained from Tiny Networks

Can we employ one neural model to efficiently dismantle many complex yet unique networks? This article provides an affirmative answer. Diverse real-world systems can be abstracted as complex networks each consisting of many functional nodes and edges. Percolation theory has indicated that removing only a few vital nodes can cause the collapse of whole network. However, finding the least number of such vital nodes is a rather challenging task for large networks due to its NP-hardness. Previous studies have proposed many centrality measures and heuristic algorithms to tackle this network dismantling (ND) problem. Different from theirs, this article tries to approach the ND task by designing a neural model which can be trained from tiny synthetic networks but will be applied for various real-world networks. It seems a discouraging mission at first sight, as network sizes and topologies are quite different across distinct real-world networks. Nonetheless, this article initiates insightful efforts of designing and training a neural influence ranking model (NIRM). Experiments on fifteen real-world networks validate its effectiveness for its mostly requiring fewer vital nodes to dismantle a network, compared with the state-of-the-art competitors. The key to its success lies in that our NIRM can efficiently encode both local structural and global topological signals for ranking nodes, in addition to our innovative labelling method in training dataset construction.

cs.SI↗

arXiv2013

Correlated diffusion of colloidal particles near a liquid-liquid interface

Optical microscopy and multi-particle tracking are used to investigate the cross-correlated diffusion of quasi two-dimensional (2D) colloidal particles near an oil-water interface. It is shown that the effect of the interface on correlated diffusion is asymmetric. Along the line joining the centers of particles, the amplitude of correlated diffusion coefficient ${D}_{\|}(r)$ is enhanced by the interface, while the decay rate of ${D}_{\|}(r)$ is hardly affected. At the direction perpendicular to the line, the decay rate of ${D}_{\bot}(r)$ is enhanced at short inter-particle separation $r$. This enhancing effect fades at the long $r$. In addition, both $D_{\|}(r)$ and $D_{\bot}(r)$ are independent of the colloidal area fraction $n$ at long $r$, which indicates that the hydrodynamic interactions (HIs) among the particles are dominated by that through the surrounding fluid at this region. However, at short $r$, $D_{\bot}(r)$ is dependent on $n$, which suggests the HIs are more contributed from the 2D particle monolayer self.

cond-mat.soft↗