The Gradient: Perspectives on AI cover image

Marc Bellemare: Distributional Reinforcement Learning

The Gradient: Perspectives on AI

00:00

How Do You Estimate a Distribution?

In reinforcement learning normally we can do this because we know the range of scores or the range of awards we're going to get. So with a categorical approach what you end up with as a histogram. I have uniformly spaced events and I'm assigning probability to each of these. With a quantile approach we fix the probabilities so we're assigning uniform probabilities to each ofThese possible outcomes. For the quantile I'm just basically trying to fit the shape of the distribution if there's more density then I'm going to have more crowded quantiles in one space. If it's less dense I'll have less crowded quantiles and so I'm getting more of a density of these objects as

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