Announcement
Bo Dai and Yilun Du’s tutorial puts machine learning under one energy-based roof
Dai and Du frame diffusion, reinforcement learning and test-time reasoning through a shared energy-based perspective.
TLDR
In their tutorial, Bo Dai and Yilun Du break energy-based models into three choices: an energy function, an inference procedure and a learning method. They argue that diffusion models, representation learning, reinforcement learning and verifier-guided inference fit this common framework. The practical payoff they propose: new algorithms built by swapping one component while keeping the others.
Combined views
753
3 Sources, first seen ago

Comments