Baiyuan Chen

Optimal Dynamic Regret by Transformers for Non-Stationary Reinforcement Learning

Feature Learning in Attention Mechanisms Is More Compact and Stable Than in Convolution