Events

CS Colloquium: On the Benefits of Convolutional Models: a Kernel Perspective

Speaker: Alberto Bietti

Location: 60 Fifth Avenue, Room 150
Videoconference link: https://nyu.zoom.us/j/93618387006

Date: Tuesday, April 19, 2022

Many supervised learning problems involve high-dimensional data such as
images, text, or graphs. In order to make efficient use of data, it is
often useful to leverage priors in the problem at hand, such as
invariance to certain transformations or stability to small
deformations. Empirically, deep convolutional networks have been very
successful on such data, raising the question of how they are able to
capture relevant structure in these problems for efficient learning.

My work studies this question from a theoretical perspective using
kernel methods, in particular convolutional kernels. These are
constructed following similar architectural principles as convolutional
networks, are closely related to their infinite-width limits in certain
regimes, and provide good empirical performance on standard computer
vision benchmarks such as Cifar10. I will present contributions that
highlight the benefits of (deep) convolutional architectures in terms of
stability to deformations and sample complexity.