CoRecursive: Coding Stories Story: Reinforcement Learning At Facebook with Jason Gauci
7 snips
Feb 1, 2021 AI Snips
Chapters
Transcript
Episode notes
Capture The Flag AI Emergence
- Jason created a neural network AI to play capture the flag using neuro evolution techniques.
- He was amazed to see emergent behaviors like players coordinating traps and decoys naturally appear.
Explore-Exploit in Recommendations
- Recommender systems must balance showing likely liked items with exploring new ones to learn user preferences.
- This explore-exploit dynamic is essential for improving recommendations beyond just ranking history.
Early Facebook Reinforcement Learning
- Facebook asked Jason to solve how to use reinforcement learning for ranking their news feed posts.
- He faced initial struggles with contractors and developed the project quietly, open sourcing it to keep working after contracts ended.