Research

Reliable intelligence, from evidence to deployment.

Publications

ARXIV2026

arXiv preprint · 2026

AsynCodeBench: Benchmarking Collaboration of Asynchronous Multi-Agent Systems in Software Engineering

Kaituo Zhang, Zhen Xiong, Zhimeng Jiang, Mingyu Zhong, Zhouyuan Yuan, Zhecheng Li, Bowen Lin, Chia-Yuan Chang, Mingzhi Hu, Huazheng Wang, Ying Lin

A dependency-centric benchmark with executable checkers, ADPR, and DRS that separates collaboration quality from coding ability in asynchronous multi-agent software engineering.

PDF ↗arXiv ↗Code ↗

ARXIV2026

arXiv preprint · 2026

Are Tools All We Need? Unveiling the Tool-Use Tax in LLM Agents

Kaituo Zhang, Zhen Xiong, Mingyu Zhong, Zhimeng Jiang, Zhouyuan Yuan, Zhecheng Li, Ying Lin

Reveals when tool-calling protocols can hurt LLM reasoning and introduces a lightweight inference-time gate.

PDF ↗arXiv ↗

TMLR2026

Transactions on Machine Learning Research · 2026

A Survey on Evaluating Quality and Trustworthiness in LLM-Generated Data

Kaituo Zhang, Mingzhi Hu, Hoang Anh Duy Le, Fariha Kabir Torsha, Zhimeng Jiang, Minh Khai Bui, Chia-Yuan Chang, Yu-Neng Chuang, Zhen Xiong, Ying Lin, Guanchu Wang, Na Zou

A metric-oriented framework for evaluating the quality and trustworthiness of LLM-generated data across modalities.

PDF ↗Code ↗

EMNLP2026

Findings of EMNLP · 2026

Cleansing the Artificial Mind: A Self-Reflective Detoxification Framework for Large Language Models

Kaituo Zhang, Zhimeng Jiang, Na Zou

A self-reflective framework that lets language models detect and correct toxic generations without external modules.

PDF ↗Code ↗

ESWA2025

Expert Systems with Applications · 2025

Robust Outlier Detection Method Based on Local Entropy and Global Density

Kaituo Zhang, Bingyang Zhang, Wei Huang, Hua Gao, Ning Xu, Rongchun Wan

EDROD combines local Shannon entropy and global density to robustly detect both point and cluster anomalies.

Paper ↗Code ↗

Research areas

Trustworthy language models

I investigate how language models represent uncertainty, preserve faithfulness, and behave fairly across people and contexts.

Synthetic data evaluation

My work studies how generated datasets should be measured—not only for resemblance, but for utility, reliability, privacy, and downstream risk.

Anomaly detection

I develop unsupervised methods that remain useful in high-dimensional settings without costly parameter tuning.