Reinforcement learning (RL) is an area of machine learning focused on teaching agents a complex relationship between its action and behavior, and maximizing a reward after a duration in an environment. The agent can be a game avatar, recommender system, notification bot, or variety of other systems that make decisions. The reward could be points in a game, or more engagement on a website. Facebook uses RL in different ways, with one example being when to let page owners know how their pages are performing.
Today, we are pleased to announce RL_Caffe2 (https://github.com/caffe2/reinforcement-learning-models), a set of RL libraries built on the Caffe2 platform. Sharing an open-source fork of our Caffe2 RL framework allows us to give back to the community and also collaborate with other institutions as RL finds more applications in industry.