Combining policy gradient and Q-learning — mentioned by indescions_2017

Sun, 22 Oct 2017 22:49:56 GMT

DeepMind's Combining Policy Gradient and Q-Learning provides good historical background: https://arxiv.org/pdf/1611.01626.pdf And archived videos from Berkeley's Deep RL Bootcamp is a crash course to get one up to speed: https://sites.google.com/view/deep-rl-bootcamp/lectures But consider the multi-model Q-distribution in Fig. 3. That graph, and the variance of its peaks, is the key to understanding why training Deep RL algorithms becomes so intractable in stochastic, real-world settings. Max Entropy Estimation and approximate Bellmen Equations will do very well in Atari games for example. Or…

Paper and discussion history. A mention is not an endorsement.

Combining policy gradient and Q-learning · HackerPapers discussions

Combining policy gradient and Q-learning — mentioned by indescions_2017