Research Portfolio

Publications

Research on reliable AI agents, multimodal reasoning, and the social and ethical behavior of large language models.

First Author

CUADebug: Diagnosing and Repairing Computer-Use Agent Failures

EMNLP 2026

CUADebug pipeline from failed trajectory diagnosis to debugger-guided rerollout

Weijia Zhang, Kunlun Zhu, Zeyi Liu, Yinting Chen, Tianyi Ma, Jiateng Liu, Jiaxun Zhang, Bingxuan Li, Pan Lu, Xiangru Tang, Heng Ji, and Jiaxuan You

A framework for diagnosing and repairing computer-use agent failures with a CUA-specific error taxonomy, benchmark, and tool-augmented debugger.

Computer-Use Agents
Failure Diagnosis
Self-Evolution
Agent Memory

ACL Oral

Understanding and Mitigating the Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks

ACL 2026, Oral

Bias inheritance research pipeline from synthetic data to downstream mitigation

Miaomiao Li, Hao Chen, Yang Wang, Tingyuan Zhu, Weijia Zhang, Kaijie Zhu, Kam-Fai Wong, and Jindong Wang

A systematic study of how bias can be inherited through LLM-generated synthetic data and how mitigation strategies behave across tasks.

Bias Inheritance
Synthetic Data
Data Augmentation
Mitigation

EMNLP System Demo

AgentDebugX: An Open-Source Toolkit for Failure Observability, Attribution, and Recovery in LLM Agents

EMNLP 2026 (System Demonstration)

AgentDebugX framework for detecting, attributing, recovering, and rerunning failed agent trajectories

Kunlun Zhu, Xuyan Ye, Zhiguang Han, Yuchen Zhao, Bingxuan Li, Weijia Zhang, Muxin Tian, Xiangru Tang, Pan Lu, James Zou, Jiaxuan You, and Heng Ji

An open-source debugging framework that turns agent trajectories into auditable failure detection, root-cause attribution, recovery, and verifiable reruns.

Agent Debugging
Failure Attribution
Recovery
Agent Trajectories

First Author · Under Review

How Much Vision Does Multimodal Reasoning Need? Vision-Stripping for Multimodal Benchmarks

Under review at NeurIPS 2026

Vision-stripping levels for multimodal benchmark reasoning

Weijia Zhang, Zijia Liu, Tianyi Zhang, Ruiqi Chen, Lian Zhang, Haoru Li, Haoqi Chen, and Jiaxuan You

Vision-stripping for multimodal benchmarks and multimodal reasoning.

Multimodal Reasoning
Vision Stripping
Benchmarks
VLM Evaluation

First Author · Preprint

Every Act Has Its Price: Compressed Moral Composition in Frontier LLMs

arXiv:2606.11232

Single-scene moral calibration and composite judgment framework

Weijia Zhang*, Ruiqi Chen*, Yunze Xiao*, and Weihao Xuan

A study of compressed moral composition in frontier LLMs.

LLM Ethics
Moral Reasoning
Frontier Models
Evaluation

ICML Workshop

Where LLM Agents Fail and How They Can Learn From Failures

Failure Modes in Agentic AI Workshop, ICML 2026

AgentDebug robot representing memory, planning, reflection, and action

Kunlun Zhu, Zijia Liu, Bingxuan Li, Muxin Tian, Yingxuan Yang, Jiaxun Zhang, Pengrui Han, Qipeng Xie, Fuyang Cui, Weijia Zhang, Xiaoteng Ma, Xiaodong Yu, Gowtham Ramesh, Jialian Wu, Zicheng Liu, Pan Lu, James Zou, and Jiaxuan You

A study of LLM agent failures and how agents can learn from failed trajectories.

LLM Agents
Failure Modes
Learning from Failure
Agent Evaluation

Co-first Author · Under Review

SeeingEye: Agentic Information Flow Unlocks Multimodal Reasoning in Text-only LLMs

arXiv:2510.25092 · AAAI 2026 under review

SeeingEye translator and text reasoning agent information flow

Weijia Zhang*, Zijia Liu*, Haoru Li*, Haoqi Chen*, and Jiaxuan You

Agentic information flow for unlocking multimodal reasoning in text-only LLMs.

Agentic Information Flow
Text-only LLMs
Multimodal Reasoning
Tool Use

Co-first Author

Artificial Leviathan: Exploring Social Evolution of LLM Agents Through the Lens of Hobbesian Social Contract Theory

The First Workshop on AI Behavioral Science, ACM SIGKDD 2024

Artificial Leviathan sandbox society simulation environment

Gordon Dai*, Weijia Zhang*, Jinhan Li, Siqi Yang, Srihas Rao, Arthur Caetano, and Misha Sra

A simulated LLM agent society for studying emergent social contracts through Hobbesian social contract theory.

Multi-Agent Systems
Social Simulation
LLM Agents
AI Behavior