Max Schwarzer
TalkRL: The Reinforcement Learning Podcast
00:00
The Evolution of Self-Predictive Representation Learning in Machine Learning
Arvind Srinivas: Can you comment on how is our understanding of what makes for a good self-predictive representation for RL? Is that our understanding of that grown a lot over these generations of agents? Yeah, absolutely. And it's funny you bring up Curl, because that was sort of the focus of the deep RL at the edge of the statistical precipice paper that we talked about earlier. He says there has been a shift away from classical computer vision and style self-supervised learning towards things that are more RL-tuned.
Transcript
Play full episode
