Research map / Label-efficient and robust learning / Self-supervised and contrastive
A critical analysis of self-supervision, or what we can learn from a single image
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 |
|---|---|---|---|
| Train BiGAN on one source image with strong augmentation instead of the full ImageNet set | BiGAN · ILSVRC-12 conv1 linear probe | accuracy: 17.7 -> 20.4 | Helped |
| Train BiGAN on one source image with strong augmentation instead of the full ImageNet set | BiGAN · ILSVRC-12 conv3 linear probe | accuracy: 31.0 -> 33.4 | Helped |
| Train DeepCluster on one source image with strong augmentation instead of the full ImageNet set | DeepCluster · ILSVRC-12 conv2 linear probe | accuracy: 32.5 -> 31.5 | No clear effect |
| Increase source images from one to 1,000 | DeepCluster · ILSVRC-12 conv5 linear probe | accuracy: 21.0 -> 26.8 | Helped |
| Train RotNet on more source images rather than a single image | RotNet · ILSVRC-12 conv5 linear probe | accuracy: 21.9 -> 29.4 | Helped |
| Add scale augmentation (ablation) | MonoGAN · CIFAR-10 conv1 linear probe | accuracy: 57.9 -> 67.9 | Helped |
| Add jitter augmentation (ablation) | MonoGAN · CIFAR-10 conv1 linear probe | accuracy: 57.9 -> 58.9 | Helped |
| Add rotation augmentation (ablation) | MonoGAN · CIFAR-10 conv1 linear probe | accuracy: 57.9 -> 61.4 | Helped |
| Add all three augmentations: scale, rotation, and jitter (ablation) | MonoGAN · CIFAR-10 conv4 linear probe | accuracy: 47.8 -> 63.5 | Helped |
| Use DeepCluster features from a single-image pretraining run for the frozen first two convolutions | AlexNet finetuning · ImageNet LSVRC-12 | Top-1 accuracy: 42.6 -> 52.5 | Helped |
Closest papers by meaning
- Beyond Learning Features: Training a Fully-Functional Classifier with ZERO Instance-Level Labels (AAAI 2022 · similarity 0.90)
- When Does Self-Supervision Help Graph Convolutional Networks? (ICML 2020 · similarity 0.88)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.87)
- Label-Free Supervision of Neural Networks with Physics and Domain Knowledge (AAAI 2017 · similarity 0.87)
- Deep Probabilistic Supervision for Image Classification (NeurIPS 2026 · similarity 0.87)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 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.