The Pathologies of Detachment in Reinforcement Learning

Go Explorer is an algorithm that rewards players for getting to new states of the game. It works better than nothing, but it doesn't solve the game and leaves a lot of it on the table. Go Explore uses two different approaches: detachment and derailment. When we solved both of those pathologies, all of a sudden we had an algorithm called Go Explore.

Transcript

Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app