Greg Yang: Title: Feature Learning in Infinite-Width Neural Networks

Speaker: Greg Yang, Microsoft Research

Date: Monday, March 15, 2021

Time: 5:00 PM to 6:30 PM Note: all times are in the Eastern Time Zone

Public: Yes

Location: Zoom:

Event Type: Seminar

Room Description:

Host: Aleksander Madry

Contact: Greg Yang: Title: Feature Learning in Infinite-Width Neural Networks,

Relevant URL:

Speaker URL: None

Speaker Photo:

Reminders to:,,

Reminder Subject: TALK: Greg Yang: Title: Feature Learning in Infinite-Width Neural Networks

Abstract: As its width tends to infinity, a deep neural network's behavior under gradient descent can become simplified and predictable (e.g. given by the Neural Tangent Kernel (NTK)), if it is parametrized appropriately (e.g. the NTK parametrization). However, we show that the standard and NTK parametrizations of a neural network do not admit infinite-width limits that can learn representations (i.e. features), which is crucial for pretraining and transfer learning such as with BERT. We propose simple modifications to the standard parametrization to allow for feature learning in the limit. Using the *Tensor Programs* technique, we derive explicit formulas for such limits. On Word2Vec and few-shot learning on Omniglot via MAML, two canonical tasks that rely crucially on feature learning, we compute these limits exactly. We find that they outperform both NTK baselines and finite-width networks, with the latter approaching the infinite-width feature learning performance as width increases.

More generally, we classify a natural space of neural network parametrizations that generalizes standard, NTK, and Mean Field parametrizations. We show 1) any parametrization in this space either admits feature learning or has an infinite-width training dynamics given by kernel gradient descent, but not both; 2) any such infinite-width limit can be computed using the Tensor Programs technique.

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Deborah Goodwin Email at Tuesday, March 09, 2021 at 5:07 PM.