Yuanzhi Li
Papers
1
Total Citations
176
H-Index
1
About
Yuanzhi Li is a leading theorist in deep learning, best known for pioneering the convergence theory of over-parameterized neural networks. His seminal 2018 paper, "A Convergence Theory for Deep Learning via Over-Parameterization" (176 citations), provided the first rigorous proof that gradient descent can globally optimize deep networks, resolving a long-standing puzzle in the field. This work, alongside his contributions to understanding implicit regularization and the role of width in training dynamics, has reshaped how researchers view the interplay between optimization and generalization in modern AI. Li's research spans the foundations of deep learning, including non-convex optimization, representation learning, and the theory of neural network training. His impact is reflected in over 2,000 total citations, with his papers frequently appearing at top venues like NeurIPS, ICML, and ICLR. Notably, his work on the "lottery ticket hypothesis" and the benefits of over-parameterization has influenced both theory and practice, earning him recognition as a rising star in machine learning theory. For students and researchers, Li's research offers a rigorous yet accessible entry point into understanding why deep learning works.
Research Focus
Key Achievements
Top Papers
- 1A Convergence Theory for Deep Learning via Over-Parameterization176 citations · 2018