Machine Learning has provided various formulations to solve problems. Reinforcement learning is the third paradigm of machine learning after supervised and unsupervised learning. Here, the objective is to develop and learn by mistakes, and unlike the other 2 paradigms, the data for this is primarily developed as they are encountered. Quoting Kaelbling, L.P. in his 1996 review paper (Reinforcement Learning: A Survey) as:

Reinforcement learning (RL) is learning by interacting with an environment

Its applications lie primarily in robotics and even games. Using RL, in 2015, AlphaGo by Google defeated the world champion in Go which has an unlimited number of possibilities. This post describes the basics of RL with an example of how to train a mountain car to reach the top of the hill (I’ve added GitHub code link in the end).

#value-iteration #reinforcement-learning #q-learning #machine-learning