Reinforcement Learning

about | blog | config | notes | github

rl_agent_environment_loop.png

Image taken from Wikipedia

Many of these notes were created from [BROKEN LINK: 4de594f7-24f5-4de9-9a8d-9d227682a490]

1. What is Reinforcement Learning?

Reinforcement Learning (RL) is study of the science of decision making. The main differences of RL from other Machine Learning paradigms is that there is no supervisor to tell the algorithm what is right or wrong, we only have the reward. This reward signal is also not instantaneous. The other key difference is that the agent is actively changing the data that it sees (it is i.i.d.).

2. Example of Reinforcement Learning Problems

  • Stunt Manoeuvers in a Helicoptor
  • Defeat the world champion at Backgammon or Go
  • Make a humanoid robot walk
  • Play atari games better than humans

A lot of RL is generally applied to games, but it doesn't have to be. Games are generally very difficult forms of environments.

3. Fundamental Concepts

Created: 2021-11-13

Emacs 26.1 (Org mode 9.5)