Research map / Label-efficient and robust learning / Self-supervised and contrastive
Semi-Supervised Learning Made Simple With Self-Supervised Clustering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 self-supervised cluster targets with the joint supervised and clustering objective that conditions shared prototypes on class labels (ablation) | Suave · ImageNet-1k, 10% and 1% labels | Acc@1: 70.2 -> 75.0; 53.9 -> 66.2 | Helped |
| Use joint training with supervised contrastive loss rather than the shared cross-entropy objective (ablation) | SwAV+CT · ImageNet-1k, 10% labels | Acc@1: 70.8 vs 75.0 | Hurt |
| Use semi-supervised fine-tuning rather than supervised fine-tuning (ablation) | Suave · ImageNet-1k, 10% labels | Acc@1: 74.8 -> 75.0 | Helped |
| Use semi-supervised fine-tuning rather than supervised fine-tuning (ablation) | Suave · ImageNet-1k, 1% labels | Acc@1: 64.8 -> 66.2 | Helped |
| Evaluate projector layer 1 rather than the backbone representation (ablation) | Suave · ImageNet-1k, 1% labels | Acc@1: 65.2 -> 66.2 | Helped |
| Evaluate projector layer 2 rather than projector layer 1 (ablation) | Suave · ImageNet-1k, 1% labels | Acc@1: 66.2 -> 65.1 | Hurt |
| Initialize Suave with a stronger, longer-trained SwAV checkpoint (ablation) | Suave · ImageNet-1k, 10% and 1% labels | Acc@1: 73.6 -> 75.0; 63.8 -> 66.2 | Helped |
| Apply the proposed semi-supervised clustering framework to SwAV | Suave · CIFAR100, 400, 2500, and 10000 labels | Acc@1: 62.2 -> 64.6; 75.7 -> 77.0; 80.1 -> 81.6 | Helped |
| Apply the proposed semi-supervised clustering framework to DINO | Daino · ImageNet-1k, 10% labels, ViT-S/16 | Acc@1: 72.2 -> 76.6 | Helped |
| Continue Suave training using the semi-supervised objective on CIFAR100 (ablation) | Suave · CIFAR100 | Acc@1 | No clear effect |
Closest papers by meaning
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.90)
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.90)
- Self-labelling via simultaneous clustering and representation learning (ICLR 2020 · similarity 0.89)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.89)
- Partial Multi-View Clustering via Self-Supervised Network (AAAI 2024 · similarity 0.88)
- Collapse-Proof Non-Contrastive Self-Supervised Learning (ICML 2025 · 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.