A Caltech Library Service

Deciding How to Decide: Dynamic Routing in Artificial Neural Networks

McGill, Mason and Perona, Pietro (2017) Deciding How to Decide: Dynamic Routing in Artificial Neural Networks. Proceedings of Machine Learning Research, 70 . pp. 2363-2372. ISSN 1938-7228.

[img] PDF - Published Version
See Usage Policy.

[img] Archive (ZIP) - Supplemental Material
See Usage Policy.

[img] PDF - Accepted Version
See Usage Policy.


Use this Persistent URL to link to this item:


We propose and systematically evaluate three strategies for training dynamically-routed artificial neural networks: graphs of learned transformations through which different input signals may take different paths. Though some approaches have advantages over others, the resulting networks are often qualitatively similar. We find that, in dynamically-routed networks trained to classify images, layers and branches become specialized to process distinct categories of images. Additionally, given a fixed computational budget, dynamically-routed networks tend to perform better than comparable statically-routed networks.

Item Type:Article
Related URLs:
URLURL TypeDescription
Perona, Pietro0000-0002-7583-5809
Additional Information:© 2017 by the author(s). This work was funded by a generous grant from Google Inc. We would also like to thank Krzysztof Chalupka, Cristina Segalin, and Oisin Mac Aodha for their thoughtful comments.
Funding AgencyGrant Number
Record Number:CaltechAUTHORS:20180614-120024280
Persistent URL:
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:87106
Deposited By: Caroline Murphy
Deposited On:14 Jun 2018 20:53
Last Modified:27 Feb 2020 23:30

Repository Staff Only: item control page