Research map / Architectures and efficiency / CNN design
Collaborative Layer-Wise Discriminative Learning in Deep Neural Networks
Xiaojie Jin, Yunpeng Chen, Jian Dong, Jiashi Feng, Shuicheng Yan
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 |
|---|---|---|---|
| use CLDL loss with collaborating classifiers instead of independent softmax losses (ablation) | DSN*-NIN · CIFAR-100 (without data augmentation) | error rate (%): 34.12 -> 30.40 | Improved |
| use CLDL loss with collaborating classifiers instead of independent softmax losses (ablation) | DSN*-NIN · CIFAR-100 (with data augmentation) | error rate (%): 32.95 -> 29.05 | Improved |
| add feedback from top classifiers to bottom classifiers (ablation) | CLDL--NIN · CIFAR-100 (without data augmentation) | error rate (%): 31.27 -> 30.40 | Improved |
| add feedback from top classifiers to bottom classifiers (ablation) | CLDL--NIN · CIFAR-100 (with data augmentation) | error rate (%): 30.41 -> 29.05 | Improved |
| train NIN with CLDL | NIN* · CIFAR-100 (without data augmentation) | error rate (%): 35.96 -> 30.40 | Improved |
| train NIN with CLDL | NIN* · CIFAR-100 (with data augmentation) | error rate (%): 32.75 -> 29.05 | Improved |
| train NIN with CLDL instead of DSN | DSN · MNIST | error rate (%): 0.39 -> 0.28 | Improved |
| train GoogLeNet with CLDL | GoogLeNet · ImageNet val | top-5 error (%): 11.1 -> 10.21 | Improved |
| fine-tune VGGNet with CLDL | VGGNet_ft · MIT67 | top-1 accuracy (%): 83.10 -> 84.69 | Improved |
| fine-tune VGGNet with CLDL | VGGNet_ft · SUN397 | top-1 accuracy (%): 68.47 -> 70.40 | Improved |
| fine-tune VGGNet with CLDL | VGGNet_ft · Places205 | top-5 (%) as printed: 87.60 -> 88.67 | Improved |
| increase number of collaborative classifiers from 1 to 3 then to 4 or 5 (ablation) | CLDL-NIN · CIFAR-100 (with data augmentation) | accuracy | Mixed |
Closest papers by meaning
- Collaborative Learning for Deep Neural Networks (NeurIPS 2018 · similarity 0.88)
- Cross-Convolutional-Layer Pooling for Image Recognition (TPAMI 2016 · similarity 0.87)
- Network of Experts for Large-Scale Image Categorization (ECCV 2016 · similarity 0.87)
- Deep Discriminative CNN with Temporal Ensembling for Ambiguously-Labeled Image Classification (AAAI 2020 · similarity 0.87)
- Greedy Layerwise Learning Can Scale To ImageNet (ICML 2019 · similarity 0.87)
- Unsupervised Learning of Discriminative Attributes and Visual Representations (CVPR 2016 · similarity 0.86)
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.
