imitation learning

Imitation learning is a machine learning approach in which a system learns how to act from records of states and actions demonstrated by an expert, rather than from an explicitly defined reward. Behavioral cloning, which copies demonstrations through supervised learning, is its simplest form, and inverse reinforcement learning, which infers a reward function from demonstrations, is also part of the field.

It is used in areas such as robot manipulation and autonomous driving, where desired behavior is hard to express as a reward. Some researchers also view next-token pretraining of language models as imitation learning on human text.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.