Qiaosheng (Eric) Zhang 张乔生

Research Scientist Shanghai Artificial Intelligence Laboratory (上海人工智能实验室) Xuhui, Shanghai, China E-mail: ericzhang8951@gmail.com, zhangqiaosheng@pjlab.org.cn

About me

• Research Scientist at the Shanghai AI Lab (since 2022).

• Ph.D. Advisor at Shanghai Innovation Institute (since 2024).

• Adjunct Ph.D. advisor at the Shanghai Jiao Tong University and Fudan University (since 2024).

• Received B.Eng. degree (2015) and Ph.D. degree (2019) in the Department of Information Engineering, The Chinese University of Hong Kong, advised by Sidharth Jaggi and Mayank Bakshi.

• Research Fellow at National University of Singapore from 2019-2022, advised by Vincent Y. F. Tan

• Visiting student at Georgia Institute of Technology (2018), advised by Matthieu Bloch

Grants and Awards

• Excellent Young Scientists Fund (Overseas) by NSFC

• Young Scholars Award, Information Theory Society of Chinese Institute of Electronics (2024)

• Outstanding Teaching Assistant Award, Department of Information Engineering, CUHK (2019)

Ph.D./Intern Opportunites

Research

Research interests

• Reinforcement Learning (RL): Online/Offline RL, RL from Human Feedback (RLHF), Multi-agent RL

• Large Language Model (LLM): LLM reasoning, LLM Agent, LLM Safety, RLHF

• Information Theory: Covert communication, Information-theoretic security, Identification, Mismatched decoding

• Community Detection (a.k.a. clustering): Stochastic Block Model (SBM), Contextual SBM, Hypergraph SBM

Projects

• MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning [Report] [CPGD algorithm] [MM-PRM] [量子位 2025.6.23] [机器之心 2025.3.14]

• SafeWork-R1: Coevolving Safety and Intelligence under the AI-45◦ Law [Report]

Selected Publications

• MAGIC: A Co-Evolving Attacker-Defender Adversarial Game for Robust LLM Safety

  X. Wen†, Z. He†, H. Qi†, Z. Wan, Z. Ma†, Y. Wen, T. Zheng, X. Xu, C. Lu Q. Zhang

  arxiv preprint: 2602.01539

• Misclassification Rates and Privacy-Utility Trade-offs in Graph Convolutional Networks via Subsampling Stability

  Y. Zhang†, Z. Ma† Q. Zhang, Z. Wang

  ICASSP 2026

• A Beginner-Friendly Tutorial on LLM-based Agents

  Submitted to Proceedings of the IEEE (preprint: https://llm-agent-tutorial.github.io/website/).

• Capacity Region for Covert Secret Key Generation over Multiple Access Channels

  Y. Zhang, L. Zhou, Q. Zhang

  IEEE Transactions on Information Theory (accepted in part in IEEE ITW 2025)

• The Avengers: A Simple Recipe for Uniting Smaller Language Models to Challenge Proprietary Giants

  Y. Zhang, H. Li, C. Wang†, L. Chen, Q. Zhang, et al.

  AAAI 2026 (Oral)

• Robust RLHF for Human Preference with Instance-Dependent Flipping

  Y. Xu†, X. Ye, Y. Chen, Q. Zhang

  AAAI 2026

• ICL-Router: In-Context Learned Model Representations for LLM Routing

  C. Wang†, H. Li, Y. Zhang, L. Chen, J. Chen, P. Jian, Q. Zhang, S. Hu.

  AAAI 2026

• Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute

  J. Chen, Z. Xun, B. Zhou†, H. Qi†, H. Zhang†, Q. Zhang, et al.

  AAAI 2026

• Adaptive Theory of Mind for LLM-based Multi-Agent Coordination

  C. Mu†, Y. Zeng, Q. Zhang, et al.

  AAAI 2026

• Unsupervised Skill Discovery through Skill Regions Differentiation

  T. Xiao, J. Zheng, R. Yang, K. Xu, Q. Zhang, P. Liu, Z. Wang, C. Bai

  IEEE Transactions on Neural Networks and Learning Systems, 2025

• Graph Feedback Bandits on Similar Arms: With and Without Graph Structures

  H. Qi†, F. Guo, L. Zhu, Q. Zhang

  Submitted to IEEE Transactions on Information Theory

• Provably Efficient Information-Directed Sampling Algorithms for Multi-Agent Reinforcement Learning

  Q. Zhang, C. Bai, S. Hu, Z. Wang, X. Li

  Artificial Intelligence (AIJ), 2025.

• Sample-Efficient Reinforcement Learning from Human Feedback via Information-Directed Sampling

  H. Qi†, H. Yang, Q. Zhang, Z. Yang

  IEEE Transactions on Information Theory, 2025

• Graph Attention is Not Always Beneficial: A Theoretical Analysis of Graph Attention Mechanisms via Contextual Stochastic Block Models

  Z. Ma†, Q. Zhang, B. Zhou†, Y. Zhang†, S. Hu, Z. Wang

  International Conference on Machine Learning (ICML), 2025.

• ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck

  Z. Zhou, W. Feng, Q. Zhang, S. Lyu, Q. Zhao, G. Cheng

  International Conference on Machine Learning (ICML), 2025.

• Online Preference Alignment for Language Models via Count-based Exploration

  C. Bai, Y. Zhang, S. Qiu, Q. Zhang, K. Xu, X. Li

  International Conference on Learning Representations (ICLR), 2025 (Spotlight, Top 5.1% ).

• Multi-LLM-Agents Debate - Performance, Efficiency, and Scaling Challenges

  H. Zhang†, Z. Cui, Q. Zhang, S. Hu

  International Conference on Learning Representations (ICLR), Blogpost Track, 2025.

• Community Detection for Contextual-LSBM: Theoretical Limitations of Misclassification Rate and Efficient Algorithms

  D. Jin†, Y. Zhang, Q. Zhang

  IEEE International Symposium on Information Theory (ISIT), 2025

• Optimal Information Security Against Limited-View Adversaries: The Benefits of Causality and Feedback

  M. Bakshi, S. Kadhe, Q. Zhang, S. Jaggi, A. Sprintson

  IEEE Transactions on Communications, 2025

• Matrix Completion with Hypergraphs: Sharp Thresholds and Efficient Algorithms

  Z. Ma†, Q. Zhang, Z. Wang

  Learning on Graphs Conference (LoG), 2024.

• Constrained Ensemble Exploration for Unsupervised Skill Discovery

  C. Bai, R. Yang, Q. Zhang, K. Xu, Y. Chen, T. Xiao, X. Li

  International Conference on Machine Learning (ICML), 2024.

• On the Role of General Function Approximation in Offline Reinforcement Learning

  C. Mao†, Q. Zhang, Z. Wang, X. Li

  International Conference on Learning Representations (ICLR), 2024. (Spotlight, Top 5% )

• Enhancing Covert Communication in OOK Schemes by Phase Deflection

  X. Ji, R. Zhu, Q. Zhang, C. Li, D. Cao

  IEEE Transactions on Information Forensic and Security, 2024.

• Ensemble Successor Representations for Task Generalization in Offline-to-Online Reinforcement Learning

  C. Wang, X. Yu, C. Bai, Q. Zhang, Z. Wang

  SCIENCE CHINA Information Sciences, 2024.

• Exact Recovery in the General Hypergraph Stochastic Block Model

  Q. Zhang, V. Y. F. Tan

  IEEE Transactions on Information Theory, 2023.

• Covert Communication with Mismatched Decoders

  Q. Zhang, V. Y. F. Tan

  IEEE Transactions on Information Theory, 2023. (accepted in part in IEEE ISIT 2022)

• Optimal Information Security Against Limited-view Adversaries: Beyond MDS Codes

  Q. Zhang, S. Kadhe, M. Bakshi, S. Jaggi, A. Sprintson

  IEEE Transactions on Communications, 2023. (accepted in part in IEEE ISIT 2015 and IEEE ITW 2015)

• Covert Communication Gains from Adversary’s Uncertainty of Phase Angles

  S. Qiao, D. Cao, Q. Zhang, Y. Xu, G. Liu

  IEEE Transactions on Information Forensic and Security, 2023.

• MC2G: An Efficient Algorithm for Matrix Completion with Social and Item Similarity Graphs

  Q. Zhang#, G. Suh#, C. Suh, V. Y. F. Tan (# indicates equal contribution)

  IEEE Transactions on Signal Processing, 2022.

• Covert Communication over Adversarially Jammed Channels

  Q. Zhang, M. Bakshi, S. Jaggi

  IEEE Transactions on Information Theory, 2021. (accepted in part in IEEE ITW 2018)

• Covert Identification over Binary-Input Discrete Memoryless Channels

  Q. Zhang, V. Y. F. Tan

  IEEE Transactions on Information Theory, 2021.

• Community Detection and Matrix Completion with Social and Item Similarity Graphs

  Q. Zhang, V. Y. F. Tan, C. Suh

  IEEE Transactions on Signal Processing, 2021. (accepted in part in IEEE ISIT 2020)

• Optimal Change-Point Detection with Training Sequences in the Large and Moderate Deviations Regimes

  H. He, Q. Zhang, V. Y. F. Tan

  IEEE Transactions on Information Theory, 2021. (accepted in part in IEEE ISITA 2020)

• Covert Communication with Polynomial Computational Complexity

  Q. Zhang, M. Bakshi, S. Jaggi

  IEEE Transactions on Information Theory, 2020. (accepted in part in IEEE ISIT 2016)

• Stealthy Communication Over Adversarially Jammed Multipath Networks

  J. Song†, Q. Zhang, S. Kadhe, M. Bakshi, S. Jaggi

  IEEE Transactions on Communications, 2020. (accepted in part in IEEE ISIT 2018)

• Undetectable Radios: Covert Communication under Spectral Mask Constraints

  Q. Zhang, M. Bloch, M. Bakshi, S. Jaggi

  IEEE International Symposium on Information Theory (ISIT), 2019