Distilling Reinforcement Learning into Single-Batch Datasets

Connor Wilhelm, Dan Ventura

Dataset distillation compresses a large dataset into a small synthetic dataset such that learning on the synthetic dataset approximates learning on the original. Training on the di…