We investigate the coherent gradient hypothesis and show that the coherence measurements are different on real and random data regardless of the network's initialization. We introduce "diffs," an attempt at an element-wise approximation at coherence, and investigate their properties. We study how coherence is affected by increasing the width of simple fully-connected networks. We then prune those fully-connected networks and find that sparse networks outperform dense networks with the same number of nonzero parameters. In addition, we show that it is possible to increase the performance of a sparse network by scaling the size of the dense parent network it is derived from. Finally we apply our pruning methods to ResNet50 and ViT and find that diff-based pruning can be competitive with other methods.
Identifer | oai:union.ndltd.org:BGMYU2/oai:scholarsarchive.byu.edu:etd-11354 |
Date | 29 April 2024 |
Creators | Yauney, Zachary |
Publisher | BYU ScholarsArchive |
Source Sets | Brigham Young University |
Detected Language | English |
Type | text |
Format | application/pdf |
Source | Theses and Dissertations |
Rights | https://lib.byu.edu/about/copyright/ |
Page generated in 0.0012 seconds