John Langford • 7/21/2020

HOMER: Provable Exploration in Reinforcement Learning

This article explains the HOMER algorithm for reinforcement learning, presented at ICML 2020. It addresses three core RL challenges: global exploration, decoding latent dynamics from rich observations, and optimizing a reward function. The post breaks down the mathematically heavy paper, using the 'combination lock' problem to illustrate how HOMER achieves provable, efficient learning in complex environments with high-dimensional observations.

0 comments

#Reinforcement Learning #Latent Dynamics #Exploration