Deep Reinforcement Learning : a review

Abstract - Since the beginning of the years 2010, achievements in reinforcement learning haven’t stop to emerge. Particularly, a lot of new algorithms derived from the merge between Deep Learning and Reinforcement Learning have proved their effectiveness. The biggest demonstrations that have been effectuated were on Atari Games and Dota2 thanks to algorithms developed respectively by DeepMind and Open AI. Indeed, in both cases, human being were defeated. This paper tried to give the explanations behind these breakthroughs.

Key words - Reinforcement Learning, Value Iteration, Policy Iteration, Q-learning, Deep Reinforcement Learning, Convolutional Neural Network, Deep Q-Network, Proximal Policy Optimization, Policy Update.

View my paper

Alexandre Berthet
PhD Student - Image Processing and Deep Learning

My research interests include Digital Image Forensics, Features Extraction, Classification, Convolutional Neural Networks and Deep Learning.