Speaker
Zhengkang Zhang
(University of Utah)
Description
I will present some recent progress at the intersection between machine learning and field theories, highlighting Feynman diagram methods for neural network correlators and neural scaling laws.
First, building on a correspondence between neural network ensembles and statistical field theories, I will introduce a diagrammatic framework to calculate neural network correlators in the large-width expansion and study RG flow and criticality. Then, I will show how large-N field theory methods can be used to solve a model of neural scaling laws.
Based on 2305.02334 and work to appear.
Author
Zhengkang Zhang
(University of Utah)
