114 0

REGULARIZING ACTIVATIONS IN NEURAL NETWORKS VIA DISTRIBUTION MATCHING WITH THE WASSERSTEIN METRIC

Title
REGULARIZING ACTIVATIONS IN NEURAL NETWORKS VIA DISTRIBUTION MATCHING WITH THE WASSERSTEIN METRIC
Author
김병훈
Keywords
regularization; Wasserstein metric; deep learning
Issue Date
2020-04
Publisher
ARXIV
Citation
International Conference on Learning Representations, Page. 1-13
Abstract
Regularization and normalization have become indispensable components in training deep neural networks, resulting in faster training and improved generalization performance. We propose the projected error function regularization loss (PER) that encourages activations to follow the standard normal distribution. PER randomly projects activations onto one-dimensional space and computes the regularization loss in the projected space. PER is similar to the Pseudo-Huber loss in the projected space, thus taking advantage of both and regularization losses. Besides, PER can capture the interaction between hidden units by projection vector drawn from a unit sphere. By doing so, PER minimizes the upper bound of the Wasserstein distance of order one between an empirical distribution of activations and the standard normal distribution. To the best of the authors' knowledge, this is the first work to regularize activations via distribution matching in the probability distribution space. We evaluate the proposed method on the image classification task and the word-level language modeling task.
URI
https://arxiv.org/abs/2002.05366https://repository.hanyang.ac.kr/handle/20.500.11754/164478
Appears in Collections:
COLLEGE OF ENGINEERING SCIENCES[E](공학대학) > INDUSTRIAL AND MANAGEMENT ENGINEERING(산업경영공학과) > Articles
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE