Deciding How to Decide: Dynamic Routing in Artificial Neural Networks
Abstract
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.
Cite
Text
McGill and Perona. "Deciding How to Decide: Dynamic Routing in Artificial Neural Networks." International Conference on Machine Learning, 2017.Markdown
[McGill and Perona. "Deciding How to Decide: Dynamic Routing in Artificial Neural Networks." International Conference on Machine Learning, 2017.](https://mlanthology.org/icml/2017/mcgill2017icml-deciding/)BibTeX
@inproceedings{mcgill2017icml-deciding,
title = {{Deciding How to Decide: Dynamic Routing in Artificial Neural Networks}},
author = {McGill, Mason and Perona, Pietro},
booktitle = {International Conference on Machine Learning},
year = {2017},
pages = {2363-2372},
volume = {70},
url = {https://mlanthology.org/icml/2017/mcgill2017icml-deciding/}
}