Controllable Orthogonalization in Training DNNs

Lei Huang, Li Liu, Fan Zhu, Diwen Wan, Zehuan Yuan, Bo Li, Ling Shao

Research output: Contribution to journalConference articlepeer-review

Abstract

Orthogonality is widely used for training deep neural networks (DNNs) due to its ability to maintain all singular values of the Jacobian close to 1 and reduce redundancy in representation. This paper proposes a computationally efficient and numerically stable orthogonalization method using Newton's iteration (ONI), to learn a layer-wise orthogonal weight matrix in DNNs. ONI works by iteratively stretching the singular values of a weight matrix towards 1. This property enables it to control the orthogonality of a weight matrix by its number of iterations. We show that our method improves the performance of image classification networks by effectively controlling the orthogonality to provide an optimal tradeoff between optimization benefits and representational capacity reduction. We also show that ONI stabilizes the training of generative adversarial networks (GANs) by maintaining the Lipschitz continuity of a network, similar to spectral normalization (SN), and further outperforms SN by providing controllable orthogonality.

Original languageEnglish (US)
Article number9157676
Pages (from-to)6428-6437
Number of pages10
JournalProceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
DOIs
StatePublished - 2020
Event2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020 - Virtual, Online, United States
Duration: Jun 14 2020Jun 19 2020

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition

Fingerprint

Dive into the research topics of 'Controllable Orthogonalization in Training DNNs'. Together they form a unique fingerprint.

Cite this