Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Representation Learning by Invariance Propagation
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 recursive kNN-graph propagation to discover positives instead of KNN (ablation) | InvP · ImageNet linear classification | top-1 center-crop accuracy: 57.6 -> 63.3 | Helped |
| add hard positive sampling (ablation) | InvP · ImageNet linear classification | top-1 center-crop accuracy: 60.7 -> 63.3 | Helped |
| add hard negative sampling (ablation) | InvP · ImageNet linear classification | top-1 center-crop accuracy: 61.9 -> 63.3 | Helped |
| use InvP rather than MoCo v2 pretraining | ResNet-50-MLP · ImageNet linear classification | top-1 accuracy: 71.1 -> 71.3 | Helped |
| use InvP rather than the strongest listed baseline | ResNet-50-MLP · Places205 linear classification | top-1 accuracy: 55.0 -> 53.5 | Hurt |
| use InvP rather than PCL | ResNet-50-MLP · Pascal VOC07 linear classification | top-1 accuracy: 84.0 -> 84.7 | Helped |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · ImageNet 1% labels | top-5 accuracy: 75.5 -> 78.2 | Helped |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · CIFAR10 transfer | top-1 accuracy: 97.7 -> 97.9 | Helped |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · CIFAR100 transfer | top-1 accuracy: 85.9 -> 84.7 | Hurt |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · Pascal VOC2007 transfer | 11 point mAP: 84.1 -> 85.4 | Helped |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · Caltech-101 transfer | mean per-class accuracy: 92.1 -> 92.5 | Helped |
| pretrain with InvP rather than SimCLR | ResNet-50-MLP · Stanford Cars transfer | top-1 accuracy: 91.3 -> 90.3 | Hurt |
Closest papers by meaning
- Unsupervised Representation Learning - an Invariant Risk Minimization Perspective (ICLR 2026 · similarity 0.89)
- Improving Transformation Invariance in Contrastive Representation Learning (ICLR 2021 · similarity 0.89)
- Learning Invariances in Neural Networks from Training Data (NeurIPS 2020 · similarity 0.88)
- Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases (NeurIPS 2020 · similarity 0.88)
- What Should Not Be Contrastive in Contrastive Learning (ICLR 2021 · similarity 0.88)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · 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.