Research map / Label-efficient and robust learning / Semi-supervised learning
Self-Supervised Wasserstein Pseudo-Labeling for Semi-Supervised Image Classification
Fariborz Taherkhani, Ali Dabouei, Sobhan Soleymani, Jeremy Dawson, Nasser M. Nasrabadi
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 SSWPL instead of FixMatch | WRN-28-2 · CIFAR-10 (250 labels) | test error (%): 5.07 ± 0.33 -> 4.11 ± 0.15 | Improved |
| Use SSWPL instead of FixMatch | WRN-28-2 · CIFAR-10 (4000 labels) | test error (%): 4.31 ± 0.15 -> 3.18 ± 0.09 | Improved |
| Use SSWPL instead of FixMatch | WRN-28-2 · CIFAR-100 (2500 labels) | test error (%): 2.64 ± 0.64 -> 2.71 ± 0.25 | Hurt |
| Use SSWPL instead of FixMatch | WRN-28-2 · CIFAR-100 (10000 labels) | test error (%): 2.36 ± 0.19 -> 2.27 ± 0.07 | Improved |
| Use SSWPL instead of FixMatch | WRN-28-2 · SVHN (250 labels) | test error (%): 2.64 ± 0.64 -> 6.16 ± 0.18 | Hurt |
| Use SSWPL instead of FixMatch | WRN-28-2 · SVHN (1000 labels) | test error (%): 2.36 ± 0.19 -> 2.27 ± 0.07 | Improved |
| Use SSWPL instead of FixMatch | ResNet-50 · ImageNet (10% labeled; top-1) | top-1 error (%): 28.54±0.52% -> 26.46 ± 0.44% | Improved |
| Use SSWPL instead of FixMatch | ResNet-50 · ImageNet (10% labeled; top-5) | top-5 error (%): 10.87±0.28% -> 9.14±0.26% | Improved |
| Replace Wasserstein-barycenter clustering with k-means (ablation) | SSWPL · CIFAR-10 (250 labels) | test error (%): 9.62 ± 0.47 -> 4.11 ± 0.15 | Improved |
| Replace Wasserstein-barycenter clustering with k-means (ablation) | SSWPL · CIFAR-10 (4000 labels) | test error (%): 7.74 ± 0.73 -> 3.18 ± 0.09 | Improved |
| Replace Wasserstein-barycenter clustering with k-means (ablation) | SSWPL · CIFAR-100 (2500 labels) | test error (%): 6.16 ± 0.18 -> 2.71 ± 0.25 | Improved |
| Replace Wasserstein-barycenter clustering with k-means (ablation) | SSWPL · CIFAR-100 (10000 labels) | test error (%): 4.59 ± 0.34 -> 2.27 ± 0.07 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 8 of 8 shared results
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
- Semi-Supervised Learning Made Simple With Self-Supervised Clustering (CVPR 2023 · similarity 0.89)
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.88)
- SimPLE: Similar Pseudo Label Exploitation for Semi-Supervised Classification (CVPR 2021 · similarity 0.88)
- Unsupervised Clustering using Pseudo-semi-supervised Learning (ICLR 2020 · similarity 0.87)
- Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning (AAAI 2021 · similarity 0.87)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · 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.
