Fields Medalist Wang Hong Also Published at NeurIPS

Wang Hong, a newly minted Fields Medalist, co-authored a paper at NeurIPS 2019 on low-rank matrix approximation, a fundamental task in machine learning and data analysis. The paper improved the approximation ratio bound of the Column Subset Selection (CSS) algorithm: for 1 ≤ p ≤ 2, it achieves (k+1)^(1/p); for p ≥ 2, it achieves (k+1)^(1-1/p), tighter than the previous O(k+1) bound. The key innovation was the use of the Riesz–Thorin interpolation theorem from harmonic analysis, a tool not commonly used in theoretical computer science at the time. Reviewers praised this as the main technical contribution and called the paper solid. Notably, among nearly 40 papers on Wang's homepage, this is the only one without a direct link. The paper exemplifies how pure mathematics can solve machine learning problems, and it matches NeurIPS 2026's new review guidelines for theory contributions, which value mathematical rigor and cross-disciplinary methodology.