I am a fifth-year PhD student in the computer science department of Columbia University, advised by Xi Chen and Christos Papadimitriou. Prior to that, I received my bachelor's degree from Yao Class at Tsinghua University.
I am broadly interested in theoretical computer science and machine learning. My current focus is understanding learning and intelligence through the lens of computation.
I am on the job market in the academic year of 2023 - 2024!
Research
The role of memory in learning
How much memory (or space) is required for a computer to successfully carry out a learning task? My research provides answers to fundamental learning problems
Memory-Query Tradeoffs for Randomized Convex Optimization (FOCS 2023) [pdf]
Xi Chen, Binghui Peng
This paper shows that quadratic memory is necessary if one wants the optimal query complexity for convex optimization.
Near Optimal Memory-Regret Tradeoff for Online Learning (FOCS 2023) [pdf]
Binghui Peng, Aviad Rubinstein
This paper provides a complete answer for online decision making (the experts problem).
Theoretical foundations of continual learning
How can machines learn over evolving data and achieve continual (lifelong) learning? My research provides theoretical insights this fundamental question
Memory Bounds for Continual Learning (FOCS 2022) [pdf]
Xi Chen, Christos Papadimitriou, Binghui Peng
This paper elucidates the importance of memory for learning sequentially classification tasks.
See [PP23] for reinforcement learning.
The Complexity of Dynamic Least-Squares Regression (FOCS 2023) [pdf]
Shunhua Jiang, Binghui Peng, Omri Weinstein
This paper points out re-learning from scratch might be necessary even for least-square regressions.
(To complexity theoretists: This work proves the hardness of approximation for OMv conjecture).
Learning and games
How can learning algorithms coordinate with other agents (people, machines) and reach an equilibrium? My research resolves decades of years open problems in this field
Fast swap regret minimization and applications to approximate correlated equillibria [pdf]
Binghui Peng, Aviad Rubinstein
This paper gives exponentially fast no-swap regret learning algorithm for correlated equilibria and resolves a series of opens questions in algorithmic game theory literature.
Complexity of Equilibria in First-Price Auctions under General Tie-Breaking Rules (STOC 2023)[pdf]
Xi Chen, Binghui Peng
This work makes substantial progress on the complexity of Bayesian Nash equilibrium in first-price auction.
