Hongxiang Zhang

Hi, I am Hongxiang Zhang (张鸿翔). I’m a Ph.D. student in the Department of Computer Science at Purdue University, advised by Prof. Tianyi Zhang.

I am looking for research internship positions. Feel free to contact me at hxxzhang@gmail.com.

My research focuses on reliable and efficient AI systems, with an emphasis on coding agents and agent memory. These agents must reason over large codebases and long interaction histories, where missing, irrelevant, or stale context can lead to inefficient exploration and unreliable decisions. I study how agents acquire, organize, and use task-relevant knowledge through memory construction, diagnostic feedback, attention steering, reasoning-time alignment, and adaptive decoding. I apply these methods to software engineering and security tasks, including code refinement, repository-level issue resolution, hallucination mitigation, and LLM-guided fuzzing.

News

Aug 22, 2026	Our work Attention-Aligned Reasoning for Large Language Models has been accepted to EMNLP 2026 (Findings).
May 27, 2026	Check out our new work FLARE, an efficient LLM code-refinement framework that uses lightweight fault localization to guide targeted repairs.
Apr 27, 2026	Check out our new work AGENT-RADAR, a training-free attention steering method for multi-agent systems that improves context management and inter-agent coordination.
Feb 27, 2026	Check out our new work Attention-Aligned Reasoning, a reasoning-time alignment method that keeps LLMs focused on prompts and intermediate goals through step-by-step attention alignment.
Jan 06, 2026	Our work LLAMAFUZZ: Large Language Model Enhanced Greybox Fuzzing has been accepted to AST 2026.

Selected Publications

Attention-Aligned Reasoning for Large Language Models

LLM Reasoning; Attention Alignment; Prompt Engineering

Hongxiang Zhang, Yuan Tian, and Tianyi Zhang

In Findings of the Association for Computational Linguistics: EMNLP 2026, 2026

EMNLP 2026 Findings

arXiv
Active Layer-Contrastive Decoding Reduces Hallucination in Large Language Model Generation

Hallucination Mitigation; Decoding; Reinforcement Learning

Hongxiang Zhang, Hao Chen, Muhao Chen, and 1 more author

In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025

EMNLP 2025 Main

arXiv Code Website
Enhancing Multi-Agent Communication through Attention Steering with Context Relevance

Multi-agent systems; Attention Alignment; Context Management

Hongxiang Zhang, Yuan Tian, and Tianyi Zhang

2026

Under review

arXiv
FLARE: Fine-Grained Diagnostic Feedback for LLM Code Refinement

Code Refinement; Fault Localization; Test-time Scaling

Yinsheng Yao, Hongxiang Zhang, Weixi Tong, and 1 more author

2026

Under review

arXiv
LLAMAFUZZ: Large Language Model Enhanced Greybox Fuzzing

LLM for Systems; Greybox Fuzzing; Software Testing

Hongxiang Zhang, Yuyang Rong, Yifeng He, and 1 more author

In Proceedings of the 2026 IEEE/ACM International Conference on Automation of Software Test, 2026

AST 2026

arXiv