MLG 029 Reinforcement Learning Intro
Feb 05, 2018
Click to Play Episode

Introduction to reinforcement learning concepts

Resources
Resources best viewed here
Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow (3rd Edition)
StatQuest - Machine Learning
Reinforcement Learning: An Introduction (2nd Ed.) by Sutton & Barto
UC Berkeley CS285: Deep Reinforcement Learning
Show Notes
  • RL definition: goal, rewards, actions ** Games (Atari, Chess, Go - Lee Sedol & Alpha Go) ** AI: learning, vision / speech, action / motion, planning ** Reasoning / knowledge vs model-based Deep RL? ** Reasoning / knowledge rep (+memory?) => Differential computers (https://deepmind.com/blog/differentiable-neural-computers/) ** vs supervised. Vision = supervised. Games = action. Trading can go both ways! ** Time: Credit assignment, delayed rewards, investment
  • Model-based v free ** Policy (what you do; gut reaction)
  • Value-based (Q-learning) vs Policy Gradient ** PG is direct: ML -> action ** Value-based indirect: Bellman stuff -> state/action values (Q-values) -> policy
  • Openai Gym, cartpole
  • Frameworks ** openai/baselines ** reinforceio/tensorforce ** NervanaSystems/coach ** rll/rllab

Comments temporarily disabled because Disqus started showing ads (and rough ones). I'll have to migrate the commenting system.