Inductive Bias of Neural Networks

Summary

Date: 
August 18, 2020 - 1:00pm - 2:30pm
Location: 
Zoom
About the Speaker
Name: 
Cengiz Pehlevan

Predicting a target output from training examples is unsolvable without additional assumptions about the nature of the target function. A learner’s performance depends crucially on how its internal assumptions (or inductive bias) align with the task at hand. I will present a theory that describes the inductive bias of neural networks using kernel methods and statistical mechanics, and discuss its applications to artificial and biological neural systems.