Author name cluster

Weizheng Gu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

2 author rows

ICML Conference 2025 Conference Paper

Reasoning Through Execution: Unifying Process and Outcome Rewards for Code Generation

Zhuohao Yu 0001
Weizheng Gu
Yidong Wang 0003
Xingru Jiang
Zhengran Zeng
Jindong Wang 0001
Wei Ye 0004
Shikun Zhang

Large Language Models excel at code generation yet struggle with complex programming tasks that demand sophisticated reasoning. To bridge this gap, traditional process supervision relies on learned reward models requiring costly training data and suffering from reward misalignment, while outcome supervision fails for complex tasks needing coordinated intermediate steps. We introduce O utcome R efining P rocess S upervision, which unifies process and outcome supervision by leveraging executable verification: a tree-structured search framework generates strategic alternatives, profiles execution metrics, and scores candidates via self-critique mechanisms that integrate runtime feedback with reasoning. Experiments across 5 models and 3 benchmarks show consistent gains, with 26. 9% higher correctness and 42. 2% improved code efficiency. The results demonstrate that ORPS enables LLMs to overcome local optima in code generation, suggesting a promising direction for combining verifiable outcomes with structured reasoning to tackle complex challenges.

Details

NeurIPS Conference 2025 Conference Paper

SAEMark: Steering Personalized Multilingual LLM Watermarks with Sparse Autoencoders

Zhuohao Yu
Xingru Jiang
Weizheng Gu
Yidong Wang
Qingsong Wen
Shikun Zhang
Wei Ye

Watermarking LLM-generated text is critical for content attribution and misinformation prevention, yet existing methods compromise text quality and require white-box model access with logit manipulation or training, which exclude API-based models and multilingual scenarios. We propose SAEMark, an inference-time framework for multi-bit watermarking that embeds personalized information through feature-based rejection sampling, fundamentally different from logit-based or rewriting-based approaches: we do not modify model outputs directly and require only black-box access, while naturally supporting multi-bit message embedding and generalizing across diverse languages and domains. We instantiate the framework using Sparse Autoencoders as deterministic feature extractors and provide theoretical worst-case analysis relating watermark accuracy to computational budget. Experiments across 4 datasets demonstrate strong watermarking performance on English, Chinese, and code while preserving text quality. SAEMark establishes a new paradigm for scalable, quality-preserving watermarks that work seamlessly with closed-source LLMs across languages and domains.

PDF Details

ECAI Conference 2024 Conference Paper

DINEX: Interpretable NLP via Diversified Natural Language Explanations

Xugang Zhou
Jindian Su
Weizheng Gu

Natural Language Explanations (NLE) are becoming increasingly important in Interpretable Natural Language Processing, which can clarify the reasoning process and improve performance. It is essential to utilize human-authored gold explanations to improve the quality of explanations produced by the generator. However, the gold standard explanations within the same dataset are produced by a fixed group of annotators, leading to a more homogeneous perspective and style. This has inspired us to improve the quality of generated explanations by enhancing the diversity of the training set. Based on this, we introduce DINEX, a two-stage framework comprising a diversified explanation generator and an explanation-aware predictor, suitable for any task related to NLE. The first stage of DINEX augments the generator’s training set through two approaches: Semantic Similarity Sampling (SSS) and Structural Variety Generation (SVG). This enables the generator to learn how to produce NLE from diverse perspectives and styles. In the second stage, DINEX improves the predictor’s ability to capture the complementary semantics between contexts and explanations. It also reduces the impact of noise on predictions through dynamic perturbations. We conduct experiments on four datasets in the domains of Question Answering and Reasoning. The results show that DINEX achieves an average performance improvement of 5. 49%, establishing a new state-of-the-art on the ComVE dataset. Evaluations by human evaluators and Large Language Models (LLMs) demonstrate that DINEX-generated explanations surpass the baseline in quality across 62. 4% of the test samples.

Details