Research map / Label-efficient and robust learning / Semi-supervised learning
Constrained Mean Shift Using Distant Yet Related Neighbors for Representation Learning
K L Navaneet, Soroush Abbasi Koohpayegani, Ajinkya Tejankar, Kossar Pourahmadi, Akshayvarun Subramanya, Hamed Pirsiavash
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 |
|---|---|---|---|
| constrain NN search with neighbors of a previous augmentation (CMSF_self) instead of unconstrained MSF | MSF (256K) · ImageNet-1k SSL (200 epochs) | linear top-1 accuracy: 72.2 -> 73.0 | Improved |
| constrain NN search (CMSF_self) instead of MSF | MSF (256K) · ImageNet-1k SSL (200 epochs) | 20-NN accuracy: 65.1 -> 66.4 | Improved |
| add multi-crop inputs | CMSF_self · ImageNet-1k SSL (200 epochs) | linear top-1 accuracy: 73.0 -> 74.4 | Improved |
| add multi-crop inputs | CMSF_self · ImageNet-1k SSL (200 epochs) | 20-NN accuracy: 66.4 -> 66.2 | Hurt |
| relax constraint by raising k' from 5 to 80 (ablation) | CMSF_self · ImageNet-1k SSL | 1-NN accuracy: 63.2 -> 61.7 | Hurt |
| use SSL CMSF_self instead of MSF | MSF · Transfer, mean over 10 datasets | mean transfer accuracy: 76.3 -> 77.4 | Improved |
| use top-10 neighbors instead of top-all in supervised CMSF (ablation) | CMSF_sup · ImageNet-1k supervised | linear top-1 accuracy: 75.7 -> 76.4 | Improved |
| use top-10 neighbors instead of top-all on coarse-grained labels (ablation) | CMSF_sup · ImageNet-coarse train, ImageNet-1k val | linear top-1 accuracy: 67.0 -> 74.2 | Improved |
| use CMSF_sup top-10 instead of cross-entropy on coarse labels | Xent · ImageNet-coarse train, ImageNet-1k val | linear top-1 accuracy: 61.4 -> 74.2 | Improved |
| use top-10 neighbors instead of top-all under label noise (ablation) | CMSF_sup · ImageNet-100 noisy labels | accuracy | Improved |
| use CMSF over cross-entropy for fine-grained transfer | Xent (1000 ep) · Transfer, mean over 10 datasets | mean transfer accuracy: 78.9 -> 80.1 | Improved |
| add pseudo-label constraint on unlabeled data (ablation) | CMSF_semi-basic · ImageNet semi-supervised (10% labels) | top-1 accuracy: 68.6 -> 69.9 | 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.
- OBoW — OBoW: Online Bag-of-Visual-Words Generation for Self-Supervised Learning (CVPR 2021) · 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
- Contrastive Mean-Shift Learning for Generalized Category Discovery (CVPR 2024 · similarity 0.84)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.84)
- Beyond Prototypes: Semantic Anchor Regularization for Better Representation Learning (AAAI 2024 · similarity 0.83)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.83)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.83)
- Can Generative Models Improve Self-Supervised Representation Learning? (AAAI 2025 · similarity 0.83)
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.
