Reinforcement learning (machine learning)

Last revised by Andrew Murphy on 2 May 2019

Reinforcement learning is one of the main algorithms used in machine learning in the context of an agent in an environment. In each timestep, this agent takes in information from their environment and performs an action. Certain actions reward the agent. 

Reinforcement learning maximizes these rewards, a difficult task given that these rewards may be very infrequent and may not have an obvious correlation with each action. Common reinforcement learning tasks including teaching a robot to walk, a car to drive. 

Reinforcement learning can be used in image classification by implementing attention mechanisms. In such a problem, a model can choose to focus on smaller parts of a high resolution image, moving to different parts depending on what it sees, and subsequently producing a predictive label. 

ADVERTISEMENT: Supporters see fewer/no ads