Published in , 2025
This paper introduces an information-theoretic approach to machine unlearning, aimed at effectively removing the influence of features or specific training data from the model while preserving overall performance and avoiding costly retraining.
Recommended citation: Shizhou Xu. (2025). "Machine Unlearning via Information Theoretic Regularization."
Download Paper | Download Slides
Published in arXiv preprint, 2024
This paper presents novel optimality criteria for subgroup splitting (named WHOMP) in comparative experiments, such as clinical trials, social experiments, and A/B tests. Furthermore, we provide efficient algorithms to estimate the WHOMP optimal solutions. Finally, the paper offers comprehensive analytical comparisons and numerical experiments to demonstrate the significant advantages of WHOMP over traditional methods, including random partitioning, covariate-adaptive randomization, rerandomization, and anti-clustering.
Recommended citation: Shizhou Xu, Thomas Strohmer. (2024). "WHOMP: Optimizing Randomized Controlled Trials via Wasserstein Homogeneity." arXiv preprint.
Download Paper | Download Slides
Published in eScholarship, University of California, 2024
This work explores fairness in machine learning through the lens of optimal transport, offering a novel framework to address fairness in algorithmic decision-making.
Recommended citation: Shizhou Xu. (2024). "Fairness in Machine Learning via Optimal Transport." eScholarship, University of California.
Download Paper | Download Slides
Published in arXiv preprint, 2024
This paper explores the compatibility and conflict between individual fairness and (the Pareto optimality of) group fairness in machine learning.
Recommended citation: Shizhou Xu, Thomas Strohmer. (2024). "On the (In)Compatibility between Individual and Group Fairness." arXiv preprint.
Download Paper | Download Slides
Published in Journal of Machine Learning Research (JMLR), 2023
This paper introduces a mathematically provable framework for achieving fair data representation in machine learning by exploring the Pareto frontier, addressing the optimal trade-offs between group fairness and accuracy via a pre-processing approach.
Recommended citation: Shizhou Xu, Thomas Strohmer. (2023). "Fair Data Representation for Machine Learning at the Pareto Frontier." Journal of Machine Learning Research (JMLR), 24 (2023), 1-63.
Download Paper | Download Slides