Research map / Label-efficient and robust learning / Test-time adaptation
Improving Transformation Invariance in Contrastive Representation Learning
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 |
|---|---|---|---|
| Add gradient regularization | SimCLR representations · CIFAR-10 and CIFAR-100 classification | top-1 test accuracy | Helped |
| Add gradient regularization | SimCLR representations · Spirograph downstream regression | test loss | Helped |
| Add gradient regularization | SimCLR representations · CIFAR-10 with increased test-time colour distortion variance | top-1 test accuracy | Helped |
| Add gradient regularization | SimCLR representations · CIFAR-10 | test loss predicting colour transformation parameter: 0.0353 ± 0.0002 -> 0.0415 ± 0.00006 | Helped |
| Add feature averaging at test time | SimCLR representations · CIFAR-10 and CIFAR-100 | test accuracy | Helped |
| Add feature averaging at test time | SimCLR representations · Spirograph | test mean squared error | Helped |
| Add gradient regularization without feature averaging (ablation) | MoCo v2 with ResNet18 · CIFAR-100 | accuracy: 52.3% ± 0.3 -> 54.1% ± 0.1 | Helped |
| Add feature averaging to gradient-regularized MoCo v2 (ablation) | MoCo v2 with ResNet18 · CIFAR-100 | accuracy: 54.1% ± 0.1 -> 60.6% ± 0.1 | Helped |
| Add gradient regularization without feature averaging (ablation) | MoCo v2 with ResNet50 · CIFAR-100 | accuracy: 57.9% ± 0.1 -> 58.9% ± 0.2 | Helped |
| Add feature averaging to gradient-regularized MoCo v2 (ablation) | MoCo v2 with ResNet50 · CIFAR-100 | accuracy: 58.9% ± 0.2 -> 64.4% ± 0.2 | Helped |
| Increase the gradient regularization strength lambda (ablation) | Spirograph representation learning · Spirograph downstream regression | total mean squared error | Mixed |
| Increase the number of averaged test-time transformations M (ablation) | Feature-averaged representations · Linear classification or regression | expected downstream loss | Helped |
Closest papers by meaning
- What Should Not Be Contrastive in Contrastive Learning (ICLR 2021 · similarity 0.90)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.88)
- Inference via Interpolation: Contrastive Representations Provably Enable Planning and Inference (NeurIPS 2024 · similarity 0.86)
- Structuring Representation Geometry with Rotationally Equivariant Contrastive Learning (ICLR 2024 · similarity 0.86)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.86)
- Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables (NeurIPS 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.