Exploration refers to taking actions speciﬁcally to obtain more training data. If we know that given context **x** , action **a** gives us a reward of 1, we do not know whether that is the best possible reward. We may want to exploit our current policy and continue taking action **a** to be relatively sure of obtaining a reward of 1. However, we may also want to explore by trying action **a**. We do not know what will happen if we try action **d**. We hope to get a reward of 2, but we run the risk of getting a reward of 0. Either way, we at least gain some knowledge.

University of Michigan - Ann Arbor

It is very challenging in the reinforcement learning to find a perfect balance between using previously used effective action(**exploitation**) and exploring new ones (**exploration**).

In order to solve this problem, simple $\epsilon$-Greedy algorithm can be used, in which we with some small probability $\epsilon$ we would select action randomly (**exploration**) and with $1 - \epsilon$ probability use previously selected effective action(**exploitation**).  

Exploration/Exploitation trade-off

Goodfellow, I., Bengio, Y., & Courville, A. (2016). $\mathit{Deep \ Learning.}$ MIT Press. Retrieved from [www.deeplearningbook.org](https://www.deeplearningbook.org) 

Learn Before

Related