SARSA — An algorithm not named by its Inventor?! 🤯
Hi there Medium Community and random visitors alike, I will be discussing the SARSA Algorithm in this post. The post will cover the algorithm’s history and how it received its name, as well as the intuition, algorithm, and demonstration.
The History of SARSA
The SARSA algorithm was invented and introduced in a paper called “On-Line Q-Learning Using Connectionist System”. The paper was published by Rummery and Niranjan in 1994. This next part of SARSA’s history is personally my favorite portion of history for any algorithm, or at least it’s in my top 20. So the authors wanted to name the algorithm “Modified Connectionist Q-Learning” which is a bit of a mouthful. But another researcher Richard Sutton suggested SARSA. Why you might ask? It is because SARSA stands for State, Action, Reward, State, Action. This represents the parameters needed by the algorithm to work properly.
SARSA is also known as a value-based algorithm, what does it mean? It means that SARSA evaluates state-action pairs (s, a) by learning one of two different value functions: V(s) or Q(s, a) — which are used to select actions. SARSA learns the Q-value (i.e…