Score networks
The vector field of a score network
This is a random note from the Pile.
Sampling via Langevin dynamics:
\[\Large \tilde{\mathbf{x}}_t = \tilde{\mathbf{x}}_{t-1} + \frac{\epsilon}{2} \nabla_{\mathbf{x}} \log p(\tilde{\mathbf{x}}_{t-1}) + \sqrt{\epsilon} \mathbf{z}_t\]
This post is licensed under CC BY 4.0 by the author.