Facebook's Open-Source Reinforcement Learning Platform – A Deep Dive

xaviergeerinck.com

118 points·thebillkidy··10 comments

Also unrelated, but hopefully not flamebait: the code blocks run off my screen on an iPad because they’re too long. Having just fixed this issue on my own website yesterday, a bit of unsolicited advice: add “overflow: auto” to them and constrain them to the standard margins for your article.

This makes me so happy to see other delevopers help!!

Why is Facebook getting into reinforcement learning? Are there applications of RL to improving social connections or extracting more data to sell to advertisers?

If you take a look at the announcement post for the framework, you can see some discussion about that: https://code.fb.com/ml-applications/horizon/

One application that they mention related (at least loosely) to social connections is deciding whether or not to deliver a notification to a user.

That makes sense. At Twitter, I redesigned the notifications system to explicitly allow for ML based decisions such as this.

Any idea on what the objective function for such a thing would be?

I'd imagine that a company which exists entirely to collect huge amounts of data might have some interest in useful ways to automatically process that data.

Most likely to make the News Feed more engaging

Reply on news.ycombinator.com