Research map / Label-efficient and robust learning / Self-supervised and contrastive
Feature Dropout: Revisiting the Role of Augmentations in Contrastive Learning
Alex Tamkin, Margalit Glasgow, Xiluo He, Noah Goodman
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 |
|---|---|---|---|
| use learned viewmaker augmentations instead of expert SimCLR augmentations | SimCLR ResNet-18 · C+Digit, CIFAR-10 transfer | accuracy: 58.8 -> 69.3 | Improved |
| use viewmaker instead of expert augmentations | SimCLR ResNet-18 · C+Shape, CIFAR-10 transfer | accuracy: 76.0 -> 79.8 | Improved |
| use viewmaker instead of expert augmentations | SimCLR ResNet-18 · C+Digit, digit transfer | accuracy: 86.7 -> 94.3 | Improved |
| use viewmaker instead of expert augmentations | SimCLR ResNet-18 · C+Letter, letter transfer | accuracy: 94.1 -> 96.9 | Improved |
| use viewmaker instead of expert augmentations | SimCLR ResNet-18 · C+Letter, CIFAR-10 transfer | accuracy: 74.8 -> 71.9 | Hurt |
| use viewmaker instead of expert augmentations | SimCLR ResNet-18 · CIFAR-10 only, CIFAR-10 transfer | accuracy: 86.2 -> 84.5 | Hurt |
| use viewmaker instead of spectral expert views | SimCLR ResNet-18 · Speech + Sound pretrain, speech accuracy | accuracy: 10.1 -> 60.8 | Improved |
| use viewmaker instead of waveform expert views | SimCLR ResNet-18 · Speech + Sound pretrain, sound accuracy | accuracy: 43.3 -> 97.0 | Improved |
| use spectral expert views instead of viewmaker | SimCLR ResNet-18 · Speech only, speech accuracy | accuracy: 92.4 -> 97.0 | Improved |
| mask viewmaker perturbation so it cannot destroy the object label (ablation) | Viewmaker · C+Shape, CIFAR-10 transfer | accuracy: 79.8 -> 26.0 | Hurt |
| mask viewmaker perturbation so it cannot destroy the object label (ablation) | Viewmaker · C+Digit, CIFAR-10 transfer | accuracy: 69.3 -> 50.7 | Hurt |
| mask viewmaker perturbation so it cannot destroy the object label (ablation) | Viewmaker · C+Letter, CIFAR-10 transfer | accuracy: 71.9 -> 23.2 | Hurt |
Closest papers by meaning
- The Power of Contrast for Feature Learning: A Theoretical Analysis (JMLR 2023 · similarity 0.89)
- An Augmentation-Aware Theory for Self-Supervised Contrastive Learning (ICML 2025 · similarity 0.87)
- How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction (ICML 2025 · similarity 0.87)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.87)
- Perfect Alignment May be Poisonous to Graph Contrastive Learning (ICML 2024 · similarity 0.87)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (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.
