Research map / Label-efficient and robust learning / Semi-supervised learning
Deep Co-Training for Semi-Supervised Image Recognition
Siyuan Qiao, Wei Shen, Zhishuai Zhang, Bo Wang, Alan Yuille
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 |
|---|---|---|---|
| increase number of views from 2 to 4 (ablation) | Deep Co-Training · CIFAR-10 (4000 labeled) | error rate (%): 9.03 -> 8.54 | Improved |
| increase number of views from 4 to 8 (ablation) | Deep Co-Training · CIFAR-10 (4000 labeled) | error rate (%): 8.54 -> 8.35 | Improved |
| increase number of views from 2 to 4 (ablation) | Deep Co-Training · SVHN (1000 labeled) | error rate (%): 3.61 -> 3.38 | Improved |
| increase number of views from 4 to 8 (ablation) | Deep Co-Training · SVHN (1000 labeled) | error rate (%): 3.38 -> 3.29 | Improved |
| add view difference loss L_dif (adversarial examples) to co-training loss (ablation) | Deep Co-Training · SVHN (1000 labeled) | validation accuracy | Improved |
| add L_dif (ablation) | Deep Co-Training · CIFAR-10 (4000 labeled) | validation accuracy | Improved |
| add data augmentation (ablation) | dual-view Deep Co-Training · CIFAR-100 (10000 labeled) | error rate (%): 38.77 -> 34.63 | Improved |
| use Deep Co-Training instead of Temporal Ensembling | CIFAR-100+ · CIFAR-100+ (10000 labeled, with augmentation) | error rate (%): 38.65 -> 34.63 | Improved |
| use Deep Co-Training instead of Mean Teacher (ResNet-18) | ImageNet 10% labeled · ImageNet val (10% labeled) | top-1 error rate (%): 49.07 -> 46.50 | Improved |
| add unlabeled data via Deep Co-Training | ResNet-18 10% supervised · ImageNet val (10% labeled) | top-1 error rate (%): 52.23 -> 46.50 | Improved |
| use dual-view Deep Co-Training instead of VAT | SVHN/CIFAR-10 · CIFAR-10 (4000 labeled) | error rate (%): 10.55 -> 9.03 | Improved |
Closest papers by meaning
- Semi-supervised Learning with Multi-Head Co-Training (AAAI 2022 · similarity 0.89)
- Semi-Supervised and Unsupervised Deep Visual Learning: A Survey (TPAMI 2022 · similarity 0.88)
- Learning by Association -- A Versatile Semi-Supervised Training Method for Neural Networks (CVPR 2017 · similarity 0.88)
- Self-paced Multi-view Co-training (JMLR 2020 · similarity 0.87)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.87)
- Semi-Supervised Deep Learning with Memory (ECCV 2018 · similarity 0.87)
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.
