I am an Assistant Professor in the Department of Statistics
and the Department of Electrical Engineering and Computer Sciences
at UC Berkeley. In June 2020, I received my Ph. D. from Stanford, where I was advised by Andrea Montanari
My research is motivated by data science
, and lies at the intersection of statistics
, machine learning
, information theory
, and computer science
. My current research interests include high dimensional probability, theory of deep learning , and theory of reinforcement learning.
Here is my Google Scholar page and my Research Blog.
Office: Evans 387.
Multi-agent reinforcement learning
- Efficient Φ-Regret Minimization in Extensive-Form Games via Online Mirror Descent. [pdf]
Yu Bai, Chi Jin, Song Mei, Ziang Song, and Tiancheng Yu.
- Sample-Efficient Learning of Correlated Equilibria in
Extensive-Form Games. [pdf] Ziang Song, Song Mei, and Yu Bai.
- Near-Optimal Learning of Extensive-Form Games with Imperfect Information. [pdf] Yu Bai, Chi Jin, Song Mei, and Tiancheng Yu.
- When Can We Learn General-Sum Markov Games with a Large Number of Players Sample-Efficiently? [pdf] Ziang Song, Song Mei, and Yu Bai.
Our goal is to develop efficient algorithms for learning equilibria in multi-player games. Applications include real-world games such as Poker, Bridge, and Go, as well as economic systems such as auction and marketing.
Deep learning and invariances
- Learning with convolution and pooling operations in kernel methods. [pdf] Theodor Misiakiewicz and Song Mei.
- The Three Stages of Learning Dynamics in High-Dimensional Kernel Methods. [pdf] Nikhil Ghosh, Song Mei, and Bin Yu.
- Learning with invariances in random features and kernel models. [pdf] Song Mei, Theodor Misiakiewicz, and Andrea Montanari.
- Generalization error of random features and kernel methods: hypercontractivity and kernel matrix concentration. [pdf] Song Mei, Theodor Misiakiewicz, and Andrea Montanari.
Our goal is to theoretically investigate the interplay between convolution structures in neural networks and the image distribution. We explained the mechanism of why convolutional neural networks (kernels) are suitable for image datasets.
High dimensional statistics
- Local convexity of the TAP free energy and AMP convergence for Z2-synchronization. [pdf] Michael Celentano, Zhou Fan, and Song Mei.
- Performance and limitations of the QAOA at constant levels on large sparse hypergraphs and spin glass models. [pdf] Joao Basso, David Gamarnik, Song Mei, and Leo Zhou.
We study statistical-computational gaps in high dimensional statistical models.