The Gradient: Perspectives on AI cover image

Marc Bellemare: Distributional Reinforcement Learning

The Gradient: Perspectives on AI

00:00

Using Probability Metrics in Distributional Resonance Learning (RL)

The fundamental principle is that as you repeat this process of backing up the future values you will come closer and closer to the true prediction. mathematically what we mean by this is that we are coming closer at a certain rate which is exponential as a function of the discount factor. Depending on how you measure distances we can perform the same Bellman backup operation and distributions. But now depending on how we measure things we'll see things get closer together or not close together at all. That's roughly the theory of contraction mapping for distributional reinforcement learning.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app