OSLNet: Deep Small-Sample Classification With an Orthogonal Softmax Layer

Authors:
Publisher website
Google Scholar
Abstract
A deep neural network of multiple nonlinear layers forms a large function space, which can easily lead to overfitting when it encounters small-sample data. To mitigate overfitting in small-sample classification, learning more discriminative features from small-sample data is becoming a new trend. To this end, this paper aims to find a subspace of neural networks that can facilitate a large decision margin. Specifically, we propose the Orthogonal Softmax Layer (OSL), which makes the weight vectors in the classification layer remain orthogonal during both the training and test processes. The Rademacher complexity of a network using the OSL is only 1/K, where K is the number of classes, of that of a network using the fully connected classification layer, leading to a tighter generalization error bound. Experimental results demonstrate that the proposed OSL has better performance than the methods used for comparison on four small-sample benchmark datasets, as well as its applicability to large-sample datasets. Codes are available at: https://github.com/dongliangchang/OSLNet.
All Related Versions
Funding Information
  • National Key Research and Development Program of China (2019YFF0303300, 2019YFF0303302, 2018YFB2100500)
  • National Natural Science Foundation of China (61906080, 61763028, 61773071, 61922015, U19B2036, 61976138, 61977047)
  • National Science and Technology Major Program of the Ministry of Science and Technology (2018ZX03001031)
  • Beijing Academy of Artificial Intelligence (BAAI2020ZJ0204)
  • Beijing Nova Programme Interdisciplinary Cooperation Project (Z191100001119140)
  • Key Program of Beijing Municipal Natural Science Foundation (L172030)
  • Hong-Liu Outstanding Youth Talents Foundation of Lanzhou University of Technology
  • Shanghai Science and Technology Committee (2015F0203-000-06)
  • Shanghai Municipal Education Commission (2019-01-07-00-01-E00003)
  • BUPT Excellent Ph.D. Students Foundation

This publication has 43 references indexed in Scilit: