NTR Webinar: Deep Ensembles Perspective, Part 2

NTR organizes and hosts scientific webinars on neural networks and invites speakers from all over the world to present their recent work at the webinars.

On May 18 Dmitry Vetrov, National Research University Higher School of Economics, Moscow, Russia, led a technical Zoom webinar on Deep Ensembles Perspective. It was the continuation of the first webinar that happened a month ago and brought many attendees: Surprising Properties of Loss Function in Deep Learning, Part 1. 

About the webinar: 

In the second part of my talk we hypothysed about the reasons for the gap between theoretical predictions and empirical results observed in practice. 

Then we discussed different ways of building ensembles of deep neural networks and compared them based on their ability of uncertainty estimation. 

Finally we revealed some interesting regularities that arise when one tries to decide what is better: to train single large or several smaller networks.

Materials available:

Webinar presentation.

Article that was discussed.

Moderator and contact:NTR CEO Nick Mikhailovsky: nickm@ntrlab.com.

Leave a Reply

Your email address will not be published. Required fields are marked *