Cross Entropy versus Label Smoothing: A Neural Collapse Perspective

no code implementations • 6 Feb 2024 • Li Guo, Keith Ross, Zifan Zhao, George Andriopoulos, Shuyang Ling, Yufeng Xu, Zixuan Dong

We first show empirically that models trained with label smoothing converge faster to neural collapse solutions and attain a stronger level of neural collapse.

Pre-training with Synthetic Data Helps Offline Reinforcement Learning

no code implementations • 1 Oct 2023 • Zecheng Wang, Che Wang, Zixuan Dong, Keith Ross

Recently, it has been shown that for offline deep reinforcement learning (DRL), pre-training Decision Transformer with a large language corpus can improve downstream performance (Reid et al., 2022).

D4RL Q-Learning +1

On the Convergence of Monte Carlo UCB for Random-Length Episodic MDPs

no code implementations • 7 Sep 2022 • Zixuan Dong, Che Wang, Keith Ross

We nevertheless show that for a large class of MDPs, which includes stochastic MDPs such as blackjack and deterministic MDPs such as Go, the Q-function in MC-UCB converges almost surely to the optimal Q function.

Open-Ended Question Answering Q-Learning