Zhijian Zhou

Hi, there! I am a second-year PhD candidate at TMLR Group in the Faculty of Engineering and Information Technology, the University of Melbourne, advised by Dr. Feng Liu and Dr. Liuhua Peng. I received my Master’s degree from the School of Artificial Intelligence at Nanjing University in 2024, where I was supervised by Dr. Wei Gao and was a member of the LAMDA Group, led by Professor Zhihua Zhou. Prior to that, I received my B.Eng. degree in Transportation Engineering from Dalian University of Technology in 2021.

My research interests focus on making hypothesis testing usable for modern machine learning problems:

Single-example testing for probabilistic ML outputs. Most classical statistical tests are designed for batch samples, while many ML applications require decisions on individual examples with probabilistic outputs. This motivates my research on single-example testing methods that provide per-example statistical significance and theoretical false-alarm control.
Data-adaptive, learnable testing beyond rigid hypothesis specification. Conventional hypothesis testing requires specifying formal hypotheses in advance and then designing a test statistic, which often forces ML problems into restrictive statistical formulations and limits practical adoption. My research aims to build learnable testing frameworks in which both the test statistic and the testing procedure are constructed from data, enabling rigorous evaluation of a broader class of questions.
Certified LLM safety and evaluation. Modern LLM safety and evaluation often rely on empirical pipelines without rigorous finite-sample guarantees, particularly in adaptive and label-efficient settings. My current research focuses on certified and cost-efficient LLM evaluation and certified detection of training data contamination, aiming to develop theory-driven methods with finite-sample validity and provable error control for LLM evaluation and safety assessment.

Selected Publications

DUAL: Learning Diverse Kernels for Aggregated Two-sample and Independence Testing

Z.-J. Zhou, X.-Y. Tian, L.-H. Peng, C. Lei, A. Schrab, D. J. Sutherland, and F. Liu

In Advances in Neural Information Processing Systems 38, 2025

Bib PDF Code

@incollection{Zhou:Tian:Peng:Lei:Schrab:Sutherland:Liu2025,
  author = {Zhou, Z.-J. and Tian, X.-Y. and Peng, L.-H. and Lei, C. and Schrab, A. and Sutherland, D.~J. and Liu, F.},
  title = {{DUAL:} Learning Diverse Kernels for Aggregated Two-sample and Independence
                    Testing},
  booktitle = {Advances in Neural Information Processing Systems 38},
  publisher = {Curran Associates},
  address = {Dutchess, NY},
  year = {2025}
}

Anchor-based Maximum Discrepancy for Relative Similarity Testing

Z.-J. Zhou, L.-H. Peng, X.-Y. Tian, and F. Liu

In Advances in Neural Information Processing Systems 38, 2025

Bib PDF Code

@incollection{Zhou:Peng:Tian:Liu2025,
  author = {Zhou, Z.-J. and Peng, L.-H. and Tian, X.-Y. and Liu, F.},
  title = {Anchor-based Maximum Discrepancy for Relative Similarity Testing},
  booktitle = {Advances in Neural Information Processing Systems 38},
  publisher = {Curran Associates},
  address = {Dutchess, NY},
  year = {2025}
}

On the Exploration of Local Significant Differences for Two-Sample Test

Z.-J. Zhou, J. Ni, J.-H. Yao, and W. Gao

In Advances in Neural Information Processing Systems 36, 2023

Bib PDF Code

@incollection{Zhou:Ni:Yao:Gao2023,
  author = {Zhou, Z.-J. and Ni, J. and Yao, J.-H. and Gao, W.},
  title = {On the Exploration of Local Significant Differences for Two-Sample Test},
  booktitle = {Advances in Neural Information Processing Systems 36},
  publisher = {Curran Associates},
  address = {Dutchess, NY},
  year = {2023}
}