Research map / Label-efficient and robust learning / Self-supervised and contrastive
Chaos is a Ladder: A New Theoretical Understanding of Contrastive Learning via Augmentation Overlap
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 |
|---|---|---|---|
| Use moderate augmentation that connects samples within a class while keeping classes separate | Contrastive learning on the synthetic random augmentation graph · Synthetic random augmentation graph | test accuracy: 100\% at r=0.1; 50\% at r=1.5 | Helped |
| Reduce augmentation strength until the graph has only a few edges (ablation) | Contrastive learning · CIFAR-10 | test accuracy: 25\% test accuracy | Hurt |
| Increase augmentation strength until inter-class connections become frequent (ablation) | Contrastive learning · CIFAR-10 | test accuracy: 29\% test accuracy; 87.6\% inter-class edges | Hurt |
| Use properly chosen augmentation strength | Contrastive learning · CIFAR-10 | intra-class edges: 96.4\% edges are intra-class edges | Helped |
| Use ARC to select augmentation settings | Contrastive learning models trained with RandomResizedCrop and ColorJitter · Composed augmentation settings | Pearson correlation with downstream accuracy: ρ=0.80 | Helped |
| Use RandomResizedCrop alone rather than the other individual tested augmentation types (ablation) | Contrastive learning · Real-world datasets | downstream accuracy | Helped |
| Use ColorJitter as the individual augmentation rather than the less powerful tested augmentations (ablation) | Contrastive learning · Real-world datasets | downstream accuracy | Helped |
| Increase RandomResizedCrop augmentation strength from too weak to moderate (ablation) | Contrastive learning · Synthetic random augmentation graph | feature discriminability | Helped |
| Increase augmentation strength beyond the moderate range (ablation) | Contrastive learning · Synthetic random augmentation graph | feature separability: r=1.5 | Hurt |
| Tune ColorJitter strength to a moderate value (ablation) | Contrastive learning with default SimCLR augmentations · Real-world datasets | downstream accuracy: 0.8 | Helped |
Closest papers by meaning
- An Augmentation-Aware Theory for Self-Supervised Contrastive Learning (ICML 2025 · similarity 0.89)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.89)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.88)
- Perfect Alignment May be Poisonous to Graph Contrastive Learning (ICML 2024 · similarity 0.87)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 2024 · similarity 0.87)
- Generalization Analysis for Supervised Contrastive Representation Learning under Non-IID Settings (ICML 2025 · 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.