Research map / Adversarial robustness and security / Certified and empirical robustness
Shaping Deep Feature Space Towards Gaussian Mixture for Visual Classification
Weitao Wan, Yu Cheng, Jiansheng Chen, Tong Wu, Yuanyi Zhong, Ming–Hsuan Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace softmax loss with GM loss | ResNet-110 · CIFAR-10 | error rate (%): 6.43 -> 4.96 | Improved |
| replace softmax loss with GM loss | ResNet-50 · ImageNet val top-1 | error rate (%): 24.7 -> 23.28 | Improved |
| replace softmax loss with GM loss | 13-layer CNN · CIFAR-100 (no augmentation, C100) | error rate (%): 25.61 -> 23.04 | Improved |
| add classification margin (α>0) to GM loss (ablation) | GM loss · MNIST (100-D features) | error rate (%): 0.49 -> 0.39 | Improved |
| add classification margin (α>0) to GM loss (ablation) | GM loss · LFW face verification | accuracy (%): 98.80 -> 99.41 | Improved |
| add likelihood regularization (λ>0) (ablation) | GM loss · LFW face verification | accuracy (%): 98.88 -> 99.41 | Improved |
| increase margin α on data without augmentation (ablation) | GM loss · CIFAR-100 (no augmentation, C100) | error rate (%): 23.04 -> 23.16 | Mixed |
| increase margin α with augmentation (ablation) | GM loss · CIFAR-100 (augmentation, C100+) | error rate (%): 20.94 -> 20.76 | Improved |
| increase margin α from 0 to 0.3 with augmentation layers (ablation) | GM loss · CIFAR-100 with augmentation layers (λ=0.1) | accuracy (%): 78.62 -> 79.24 | Improved |
| add GM loss to K-density detector | ResNet-32 · CIFAR-10 white-box C&W-wb attack | detection AUC-ROC (%): 53.8 -> 84.3 | Improved |
| add GM loss to K-density detector | ResNet · CIFAR-10 black-box C&W-wb attack | detection AUC-ROC (%): 75.0 -> 94.7 | Improved |
| add GM loss to K-density detector | ResNet-32 · MNIST semi white-box BIM | detection AUC-ROC (%): 90.3 -> 99.7 | Improved |
Closest papers by meaning
- Rethinking Feature Distribution for Loss Functions in Image Classification (CVPR 2018 · similarity 0.94)
- Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness (ICLR 2020 · similarity 0.87)
- Gaussian Mixture Convolution Networks (ICLR 2022 · similarity 0.86)
- Large-Margin Softmax Loss for Convolutional Neural Networks (ICML 2016 · similarity 0.86)
- Classification of Data Generated by Gaussian Mixture Models Using Deep ReLU Networks (JMLR 2024 · similarity 0.86)
- Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
