Rethinking Feature Distribution for Loss Functions in Image Classification

被引:107
作者
Wan, Weitao [1 ]
Zhong, Yuanyi [1 ,2 ,3 ]
Li, Tianpeng [1 ]
Chen, Jiansheng [1 ]
机构
[1] Tsinghua Univ, Dept Elect Engn, Beijing, Peoples R China
[2] Univ Urbana Champaign, Dept Comp Sci, Champaign, IL USA
[3] Tsinghua Univ, Beijing, Peoples R China
来源
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) | 2018年
基金
中国国家自然科学基金;
关键词
DEEP NEURAL-NETWORKS;
D O I
10.1109/CVPR.2018.00950
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We propose a large-margin Gaussian Mixture (L-GM) loss for deep neural networks in classification tasks. Different from the softmax cross-entropy loss, our proposal is established on the assumption that the deep features of the training set follow a Gaussian Mixture distribution. By involving a classification margin and a likelihood regularization, the L-GM loss facilitates both a high classification performance and an accurate modeling of the training feature distribution. As such, the L-GM loss is superior to the softmax loss and its major variants in the sense that besides classification, it can be readily used to distinguish abnormal inputs, such as the adversarial examples, based on their features' likelihood to the training feature distribution. Extensive experiments on various recognition benchmarks like MNIST, CIFAR, ImageNet and LFW, as well as on adversarial examples demonstrate the effectiveness of our proposal.
引用
收藏
页码:9117 / 9126
页数:10
相关论文
共 37 条
  • [1] [Anonymous], INT C LEARN REPR
  • [2] [Anonymous], 2015, ARXIV PREPRINT ARXIV
  • [3] [Anonymous], 2013, INT C MACHINE LEARNI
  • [4] [Anonymous], IEEE C COMP VIS PATT
  • [5] [Anonymous], 2017, IEEE C COMP VIS PATT
  • [6] [Anonymous], 2014, ACM INT C MULTIMEDIA
  • [7] [Anonymous], 2017, P 5 INT C LEARN REPR
  • [8] [Anonymous], 2016, BMVC
  • [9] [Anonymous], 2006, Pattern Recognition and Machine Learning
  • [10] [Anonymous], 2016, CoRR