William Fang

MS Student
Computer Science, Mathematics
Boston University
email: wfang "at" bu "dot" edu

I am interested in private data analysis (differential privacy, unlearning, and synthetic data generation).

I am advised by Mark Bun in the Theoretical Computer Science research group.

- organizes graduate student and faculty talks on ML, Dynamics, Graphs at the BU SIAM seminar series

Privacy Preserving Machine Learning| Differential Privacy, Learning Theory

Differentially Private Winnow for Learning Halfspaces and Decision Lists @ TPDP 2025
Mark Bun, William Fang

We give new differentially private algorithms for the classic problems of learning large-margin halfspaces and decision lists in the online and PAC models. In the online model, we give a private analog of the influential Winnow algorithm for learning halfspaces with mistake bound polylogarithmic in the dimension and inverse polynomial in the margin. As an application, we describe how to privately learn decision lists in the online model, qualitatively matching state-of-the art non-private guarantees. We also study the problem of privately PAC learning decision lists, where we give a computationally efficient algorithm with minimal sample overhead over the best non-private algorithms.

Responsible & Safe AI | Machine Unlearning, Computer Vision, XAI

Developed a dual-objective approximate machine unlearning algorithm for Image classification (NeurIPS 2023 Machine Unlearning Challenge).
Experimented with other unlearning methods (i.e. SCalable Remembering and Unlearning unBound (SCRUB), teacher-student, negative gradient descent, and variants with relative entropy regularization) on CIFAR-10.
Graphical Models (on pause)