Zehao Wang
Third-year Ph.D. Student in Software Engineering, Tianjin University
I study reliable and self-improving LLM agents and multi-agent systems. My research focuses on how agents can make reliable decisions, coordinate effectively, diagnose failures, and continuously improve from interaction and experience.
My current interests lie at the intersection of Multi-Agent Systems, Agent Reinforcement Learning, Epistemic Reasoning, and Trustworthy AI, with an emphasis on building agent systems that can reason about their own decisions and learn better strategies over time.
News
2026.08
Our work on failure reasoning in LLM-based multi-agent systems was accepted to EMNLP 2026.
2026.05
Our work on epistemic calibration in LLM-based multi-agent planning was accepted to ICML 2026.
2026.01
Our work on illicit account detection based on user behavior sequences was accepted to WWW 2026.
2026
Continuing research on reliable and self-improving LLM agents.
Publications
LLM Agents & Multi-Agent Systems
EMNLP 2026
DCFA: Dual-view Causal Attribution for Failure Reasoning in LLM-based Multi-agent Systems.
ICML 2026
When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems.
Trustworthy AI & Risk Control
WWW 2026
Pattern-aware Illicit Account Detection based on User Behavior Sequences.
DASFAA 2026
NK-GAD: Neighbor Knowledge-Enhanced Unsupervised Graph Anomaly Detection.
MIR 2026
Reasoning-targeted Jailbreak Attacks on Large Reasoning Models via Semantic Triggers and Psychological Framing.
Machine Intelligence Research.
TIFS 2025
Enhancing GCN Robustness Against Structural Attacks via Adaptive Spectrum Filtering.
IEEE Transactions on Information Forensics and Security.
Neurocomputing 2025
A distribution feature extracting network with dual correlation for long sequence time-series forecasting.
★ Corresponding Author
TAI 2024
Representing Multi-view Time-series Graph Structures for Multivariate Long-term Time-series Forecasting.
IEEE Transactions on Artificial Intelligence.
CIKM 2024
Bots Shield Fake News: Adversarial Attack on User Engagement-based Fake News Detection.
FGCS 2023
A robust feature reinforcement framework for heterogeneous graphs neural networks.
Future Generation Computer Systems.
Neural Networks 2023
An Adversarial Time-Frequency Reconstruction Network for Unsupervised Anomaly Detection.
TETC 2022
Sepformer-based Models: More Efficient Models for Long Sequence Time-Series Forecasting.
IEEE Transactions on Emerging Topics in Computing.
Research
My research is organized around a central question:
How can LLM agents become more reliable and continuously improve from their own experience?
Reliable LLM Agents
I study agent reliability from a decision-centric and epistemic perspective, focusing on whether agents possess and appropriately use the knowledge required for reliable decisions.
- Epistemic calibration
- Decision reliability
- Reliable planning & execution
- Failure diagnosis & attribution
Self-Improving Agents
I investigate how agents can learn from interaction, experience, and failures rather than relying solely on static prompts or external supervision.
- Agent reinforcement learning
- Long-horizon trajectories
- Credit assignment
- Self-evolving agents
Trustworthy AI
I work on trustworthy AI and intelligent risk-control systems, where behavioral modeling, reasoning, and anomaly detection intersect.
- Risk reasoning & intervention
- LLM safety & robustness
- Graph & temporal anomaly detection
- Illicit-account detection
Research Projects
Reliable Multi-Agent Systems
Developing methods to improve the reliability of LLM-based multi-agent systems through epistemic calibration, failure attribution, hierarchical intervention, and structured reasoning.
Self-Evolving Agent Systems
Exploring online knowledge evolution and agent reinforcement learning to enable agents to learn useful decision knowledge from interaction and continuously improve their behavior.
AI Risk Control
Applying LLM reasoning, behavioral modeling, graph learning, and anomaly detection to intelligent risk detection and intervention in real-world systems.
Honors & Awards
- National Scholarship, Ministry of Education of China, 2023
- Huawei Scholarship, 2023
- Silver Award, China International College Students’ Innovation and Entrepreneurship Competition, International Track, 2024
Research Grants & Leadership
- Principal Investigator, General Research Project, Zhejiang Provincial Department of Education, 2022–2023
- Principal Investigator, Zhejiang Province Xinmiao Talent Program, 2023–2024
Research & Industry Experience
Huawei / FusionServer Research Institute
Research / System Architect
2025.11 – Present
- Research on reliable LLM-based multi-agent systems.
- Agent failure diagnosis, attribution, and system-level reliability.
- Multi-agent workflow design and evaluation.
- Research translation toward large-scale enterprise AI systems.
Tencent WeChat / Tencent Rhino-Bird Research Program
Student Lead — Intelligent Risk Control Research Project
2024.11 – 2025.08
- Led the overall research and technical development.
- User behavior modeling and illicit-account detection.
- Behavioral sequence analysis and anomaly detection.
- Core technical development, experimental evaluation, and project delivery.
- Project received Excellent Project Completion (Top 25%).
Education
Tianjin University
Ph.D. Student in Software Engineering
2024.09 – 2028.06
Hangzhou Dianzi University
M.S. in Computer Technology
2021.09 – 2024.06
Xidian University
B.Eng. in Communication Engineering
2016.09 – 2020.06
Research Collaboration
I am interested in research collaborations on:
- LLM agents and multi-agent systems
- Agent reinforcement learning
- Reliable and trustworthy AI
- Agent reasoning and decision making
- Self-improving and self-evolving agents
If you are interested in discussing research ideas or potential collaboration, feel free to reach out.