TalkRL: The Reinforcement Learning Podcast Sharath Chandra Raparthy
Feb 12, 2024
Sharath Chandra Raparthy, an AI Resident at FAIR at Meta, discusses in-context learning for sequential decision tasks, training models to adapt to unseen tasks and randomized environments, properties of data for in-context learning, burstiness and trajectories in transformers, and the use of G flow nets in sampling from complex distributions.
Chapters
Transcript
Episode notes
1 2 3 4 5 6
Introduction
00:00 • 3min
Generalizing to Unseen Tasks and Randomized Environments
03:05 • 4min
Related Work on Transformers for Sequential Decision Making
07:20 • 2min
Properties of Data for In Context Learning
09:09 • 4min
Burstiness and Trajectories in Transformers
12:51 • 18min
G Flow Nets: Sampling from Complex Distributions
30:59 • 10min