Latent Space: The AI Engineer Podcast · Latent.Space

[NeurIPS Best Paper] 1000 Layer Networks for Self-Supervised RL — Kevin Wang et al, Princeton

·28 min·2 clips

Kevin describes how doubling network depth with residual connections caused performance to skyrocket in one environment.

The hosts interview the authors of a NeurIPS Best Paper on training 1000-layer networks for self-supervised reinforcement learning. They discuss the breakthrough of scaling depth with architectural tweaks, how it blurs lines with self-supervised learning, and potential applications in robotics. The team highlights the method's efficiency and accessibility, running on a single GPU.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Latent Space: The AI Engineer Podcast