Zixuan is a Ph.D. student in Computer Science at NYU Shanghai and NYU Courant, being advised by Professor Keith Ross. He is broadly interested in the application and theory of reinforcement learning (RL) and deep learning, primarily focusing on enhancing the sample efficiency and generalization of deep RL algorithms. He holds a B.Sc. in Honors Mathematics and Data Science with a concentration in AI from NYU Shanghai. During his undergraduate, he worked with Professor Keith Ross on the convergence property of classic algorithms in tabular RL and Multi-armed Bandit literature. Outside of research, he enjoys practicing Kendo and cooking.
Publications
- George Andriopoulos*, Zixuan Dong*, Li Guo*, Zifan Zhao*, Keith Ross*, “The Prevalence of Neural Collapse in Neural Multivariate Regression”, Accepted by NeurIPS 2024
- Zecheng Wang*, Che Wang*, Zixuan Dong*, Keith Ross, “Pre-training with Synthetic Data Helps Offline Reinforcement Learning”, Accepted by ICLR 2024
- Li Guo, Keith Ross, Zifan Zhao, George Andriopoulos, Shuyang Ling, Yufeng Xu, Zixuan Dong, “Cross Entropy versus Label Smoothing: A Neural Collapse Perspective”, arXiv preprint, 2024
- Zixuan Dong, Che Wang, Keith W. Ross, “On the Convergence of Monte Carlo UCB for Random-Length Episodic MDPs”, arXiv preprint, 2022
- (* Equal Contribution)
Research Interests
- Reinforcement Learning (RL)
- Deep Learning
- RL with Human Feedback
- LLM Reasoning with RL