Research map / Label-efficient and robust learning / Self-supervised and contrastive
All4One: Symbiotic Neighbour Contrastive Learning via Self-Attention and Redundancy Reduction
Imanol G. Estepa, Ignacio Sarasua, Bhalaji Nagarajan, Petia Radeva
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 |
|---|---|---|---|
| add centroid objective to NNCLR objective (ablation) | All4One · CIFAR-100 | top-1 accuracy: 69.62 -> 71.02 | Improved |
| add EMA to NNCLR + centroid (ablation) | All4One · CIFAR-100 | top-1 accuracy: 71.02 -> 71.08 | No clear effect |
| add feature contrast (redundancy reduction) to NNCLR + centroid + EMA (ablation) | All4One · CIFAR-100 | top-1 accuracy: 71.08 -> 72.17 | Improved |
| add neighbour objective to centroid + feature + EMA (ablation) | All4One · CIFAR-100 | top-1 accuracy: 71.31 -> 72.17 | Improved |
| add centroid objective to NNCLR + feature + EMA (ablation) | All4One · CIFAR-100 | top-1 accuracy: 71.64 -> 72.17 | Improved |
| combine feature contrast with neighbour contrast to cut embedding dimension from 2048 to 256 (ablation) | All4One vs Barlow Twins · CIFAR-100 | top-1 accuracy: Barlow Twins 71.21 (2048) -> 62.14 (256); All4One (256) 72.17 | Improved |
| use All4One under crop-only augmentation instead of NNCLR (ablation) | All4One · CIFAR-100 (weak augmentation) | top-1 accuracy: 35.39 -> 44.7 | Improved |
| increase transformer encoder layers from 3 to 9 (ablation) | All4One · CIFAR-100 | top-1 accuracy: 72.17 -> 71.75 | Hurt |
| increase number of neighbours from 5 to 20 (ablation) | All4One · CIFAR-100 | top-1 accuracy: 72.17 -> 71.79 | Hurt |
| replace NNCLR with All4One on Pets transfer | All4One · Pets transfer | top-1 accuracy: 83.34 -> 81.57 | Hurt |
| higher NN retrieval accuracy does not imply better accuracy (ablation) | All4One · CIFAR-100 | NN retrieval vs top-1: 86.16 (v3) vs 82.16 (v6, best top-1 72.17) | No clear effect |
Later papers that compared with this method
1 later accepted paper reports All4One as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Contrastive Learning with Synthetic Positives (ECCV 2024 · better on 6 of 8 shared results)
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.
- Barlow Twins — Barlow Twins: Self-Supervised Learning via Redundancy Reduction (ICML 2021) · better on 4 of 4 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
- Revisiting Contrastive Learning through the Lens of Neighborhood Component Analysis: an Integrated Framework (ICML 2022 · similarity 0.88)
- Your Contrastive Learning Is Secretly Doing Stochastic Neighbor Embedding (ICLR 2023 · similarity 0.87)
- Self-Contrastive Learning: Single-Viewed Supervised Contrastive Framework Using Sub-network (AAAI 2023 · similarity 0.86)
- Soft Neighbors are Positive Supporters in Contrastive Visual Representation Learning (ICLR 2023 · similarity 0.86)
- Neighborhood Contrastive Learning for Novel Class Discovery (CVPR 2021 · similarity 0.86)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.84)
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.
