Research map / Label-efficient and robust learning / Knowledge distillation
Knowledge Distillation Meets Open-Set Semi-supervised Learning
Jing Yang, Xiatian Zhu, Adrian Bulat, Brais Martínez, Georgios Tzimiropoulos
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 MSE loss instead of KL loss (ablation) | SRD · CIFAR-100 (WRN40-4 to WRN16-4) | Top-1 accuracy (%): 79.04 -> 79.58 | Improved |
| use MSE loss instead of pMSE loss (ablation) | SRD · CIFAR-100 (WRN40-4 to WRN16-4) | Top-1 accuracy (%): 79.13 -> 79.58 | Improved |
| add SRD loss to regularizer R (ablation) | SRD · CIFAR-100 (WRN40-4 to WRN16-4) | Top-1 accuracy (%): 78.05 -> 79.58 | Improved |
| add regularizer R to SRD loss (ablation) | SRD · CIFAR-100 (WRN40-4 to WRN16-4) | Top-1 accuracy (%): 79.10 -> 79.58 | Improved |
| add logit-matching KD to SRD (ablation) | SRD · CIFAR-100 (WRN40-4 to WRN16-4) | Top-1 accuracy (%): 79.58 -> 79.63 | Improved |
| add open-set OOD detection to SRD (ablation) | SRD · Open-set SSL (ResNet8x4) | Top-1 accuracy (%): 76.24 -> 76.08 | No clear effect |
| add data-augmentation consistency to SRD (ablation) | SRD · Open-set SSL (ResNet8x4) | Top-1 accuracy (%): 76.24 -> 75.66 | Hurt |
| add data-augmentation consistency to KD (ablation) | KD · Open-set SSL (ResNet8x4) | Top-1 accuracy (%): 74.68 -> 72.82 | Hurt |
| add MixMatch to SRD (ablation) | SRD · Open-set SSL (ResNet8x4) | Top-1 accuracy (%): 76.24 -> 77.13 | Improved |
| add SSKD to SRD (ablation) | SRD · Open-set SSL (ResNet8x4) | Top-1 accuracy (%): 76.24 -> 76.38 | Improved |
| add unlabeled Tiny-ImageNet data (D+U vs D) (ablation) | SRD · Open-set SSL (ResNet32x4 to ResNet8x4) | Top-1 accuracy (%): 75.92 -> 76.24 | Improved |
| add unlabeled Tiny-ImageNet data (D+U vs D) (ablation) | CRD (pseudo labels) · Open-set SSL (ResNet32x4 to ResNet8x4) | Top-1 accuracy (%): 75.51 -> 73.84 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- OpenMatch — OpenMatch: Open-Set Semi-supervised Learning with Open-set Consistency Regularization (NeurIPS 2021) · better on 1 of 1 shared result
- MixMatch — MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · similarity 0.92)
- Similarity-Preserving Knowledge Distillation (ICCV 2019 · similarity 0.90)
- Contrastive Representation Distillation (ICLR 2020 · similarity 0.90)
- Student Customized Knowledge Distillation: Bridging the Gap Between Student and Teacher (ICCV 2021 · similarity 0.90)
- Knowledge Distillation Meets Self-Supervision (ECCV 2020 · similarity 0.90)
- Scaled Decoupled Distillation (CVPR 2024 · similarity 0.89)
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.
