Shizhou Xu

Shizhou Xu

Working on the interdisciplinary area between mathematics and machine learning.

Publications

You can also find my articles on my Google Scholar profile.

Journal Articles

Utility–Separation Pareto Frontier: An Information-Theoretic Characterization

Published in Under review (Journal of Machine Learning Research — JMLR), 2025

We provide an information-theoretic characterization of the utility–separation trade-off, yielding a principled Pareto frontier perspective for designing and evaluating separation-based objectives.

Recommended citation: Shizhou Xu. (2025). “Utility–Separation Pareto Frontier: An Information-Theoretic Characterization.” Under review at Journal of Machine Learning Research.
Download Paper | Download Slides

Multi-resolution Enhancement for Full Spectrum Neural Representations

Published in Under review (Nature Machine Intelligence), 2025

We develop multi-resolution enhancement strategies for full-spectrum neural representations, improving fidelity across scales with an emphasis on robust learning and generalization.

Recommended citation: Yuan Ni, Z. Chen, Shizhou Xu, C. Peng, R. Plumley, C. H. Yoon, J. Thayer, J. Turner. (2025). “Multi-resolution Enhancement for Full Spectrum Neural Representations.” Under review at Nature Machine Intelligence.
Download Paper | Download Slides

Machine Unlearning via Information-Theoretic Regularization

Published in Manuscript (available on request), 2025

We develop information-theoretic regularization principles for machine unlearning, aiming to remove targeted information while maintaining general utility and enabling principled evaluation.

Recommended citation: Shizhou Xu, Thomas Strohmer. (2025). “Machine Unlearning via Information-Theoretic Regularization.” Manuscript.
Download Paper | Download Slides

Forgetting-MarI: LLM Unlearning via Marginal Information Regularization

Published in Under review (ICLR 2026), 2025

We propose marginal-information regularization for LLM unlearning, targeting targeted forgetting with strong utility retention and practical, evaluation-driven design.

Recommended citation: Shizhou Xu, Yuan Ni, Stefan Broecker, Thomas Strohmer. (2025). “Forgetting-MarI: LLM Unlearning via Marginal Information Regularization.” Under review at ICLR 2026.
Download Paper | Download Slides

WHOMP: Improving Upon Randomized Controlled Trials via Wasserstein Homogeneity

Published in Under review (Journal of the American Statistical Association), 2024

We introduce WHOMP, a Wasserstein-homogeneity optimality principle for subgroup splitting in comparative experiments (clinical trials, social experiments, and A/B tests). The method yields interpretable criteria, efficient estimators, and strong empirical gains over random partitioning, covariate-adaptive randomization, rerandomization, and anti-clustering baselines.

Recommended citation: Shizhou Xu, Thomas Strohmer. (2024). “WHOMP: Improving Upon Randomized Controlled Trials via Wasserstein Homogeneity.” Under review at Journal of the American Statistical Association.
Download Paper | Download Slides

On the (In)Compatibility between Individual and Group Fairness

Published in Under review (SIAM Journal on Mathematics of Data Science — SIMODS), 2024

We analyze fundamental tensions between individual and group fairness notions, clarifying when they can or cannot be simultaneously satisfied and what trade-offs are unavoidable.

Recommended citation: Shizhou Xu, Thomas Strohmer. (2024). “On the (In)Compatibility between Individual and Group Fairness.” Under review at SIAM Journal on Mathematics of Data Science.
Download Paper | Download Slides