Year: 2024
Author: Leonid Berlyand, Etienne Sandier, Yitzchak Shmalo, Lei Zhang
Journal of Machine Learning, Vol. 3 (2024), Iss. 4 : pp. 347–412
Abstract
We explore the applications of random matrix theory (RMT) in the training of deep neural networks (DNNs), focusing on layer pruning that reduces the number of DNN parameters (weights). Our numerical results show that this pruning leads to a drastic reduction of parameters while not reducing the accuracy of DNNs and convolutional neural network (CNNs). Moreover, pruning the fully connected DNNs actually increases the accuracy and decreases the variance for random initializations. Our numerics indicate that this enhancement in accuracy is due to the simplification of the loss landscape. We next provide rigorous mathematical underpinning of these numerical results by proving the RMT-based Pruning Theorem. Our results offer valuable insights into the practical application of RMT for the creation of more efficient and accurate deep-learning models.
Journal Article Details
Publisher Name: Global Science Press
Language: English
DOI: https://doi.org/10.4208/jml.231220
Journal of Machine Learning, Vol. 3 (2024), Iss. 4 : pp. 347–412
Published online: 2024-01
AMS Subject Headings:
Copyright: COPYRIGHT: © Global Science Press
Pages: 66
Keywords: Deep learning Marchenko-Pastur distribution Random matrix theory Increasing accuracy Pruning.
Author Details
-
Stability of accuracy for the training of DNNs via the uniform doubling condition
Shmalo, Yitzchak
Annals of Mathematics and Artificial Intelligence, Vol. 92 (2024), Iss. 2 P.439
https://doi.org/10.1007/s10472-023-09919-1 [Citations: 0]