|
Zhijie Zheng (郑志杰) I am a junior undergraduate student at Beihang University (BUAA), supervised by Prof. Lu Sheng. I am currently a research intern at Shanghai AI Lab, advised by Prof. Jing Shao. My research focuses on incorporating human priors into AI systems to make them more reliable and efficient. I am particularly interested in reinforcement learning and multimodal reasoning. |
Visitor stats |
News
|
Selected Publications(*, †, ‡ indicates equal contributions, corresponding author, and co-leads, respectively.) |
|
StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing
Zhijie Zheng*, Yu Li*, Chen Qian, Yuqian Fu, Yanwei Fu, Lu Sheng, Jing Shao, Dongrui Liu† EMNLP 2026 project page / code / paper A 4B step-level guard for pre-execution safety checks and trajectory auditing, trained with StepGen and Balance-GRPO. |
|
When AI Agents Collude Online: Financial Fraud Risks by Collaborative LLM Agents on Social Platforms
Qibing Ren*, Zhijie Zheng*, Jiaxuan Guo, Junchi Yan, Lizhuang Ma†, Jing Shao† ICLR, 2026 arXiv / code / project page / 机器之心 First benchmark modeling the full lifecycle of multi-agent financial fraud and collusion on social platforms. |
|
Geometrically-Constrained Agent for Spatial Reasoning
Zeren Chen*, Xiaoya Lu*, Zhijie Zheng, Pengrui Li, Lehan He, Yijin Zhou, Jing Shao, Bohan Zhuang†, Lu Sheng† CVPR, 2026 arXiv / code / project page / 机器之心 A general-purpose, training-free agentic framework for spatial reasoning that bridges the semantic-to-geometric gap in VLMs. |
Technical Reports (Teamwork) |
|
AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security
AgentDoG Team (Core Contributor to Post-Training) arXiv, 2026 arXiv / code / project page / Hugging Face #1 Paper of the Day on Hugging Face A lightweight and scalable alignment framework for AI agent safety, with taxonomy-guided data synthesis and efficient SFT/RL post-training. |
|
Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report (SafeWork-F1)
Shanghai AI Lab (Core Contributor) arXiv, 2025 arXiv / blog Featured by Jack Clark (Anthropic Co-founder) Comprehensive risk assessment of frontier AI models across seven critical areas including cyber offense, biological risks, and autonomous AI R&D. |
Education
|
Experience
|
|
Template from Jon Barron. |