Research map / Label-efficient and robust learning / Self-supervised and contrastive
Your Contrastive Learning Is Secretly Doing Stochastic Neighbor Embedding
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 t-SNE-style matching and Euclidean feature space in MoCo-v2 | MoCo-v2 · ImageNet linear probe | accuracy: 67.5% -> 67.0% | No clear effect |
| Use t-SNE-style matching and Euclidean feature space in MoCo-v2 | MoCo-v2 · Domain transfer benchmarks | accuracy: 8 out of 9 datasets; 75.37 -> 77.78 average | Mixed |
| Use t-SNE-style matching and Euclidean feature space in MoCo-v2 | MoCo-v2 · OOD domain generalization benchmarks | accuracy: 55.2 -> 59.5 average | Helped |
| Use t-distribution matching and Euclidean feature space | SimCLR · CIFAR-10 | nearest-neighbor accuracy: 80.8% -> 83.9% | Helped |
| Use t-distribution matching and Euclidean feature space at feature dimension 8 | SimCLR · CIFAR-10, feature dimension 8 | nearest-neighbor accuracy: 81.7% | Helped |
| Weight positive pairs by crop IoU with τ′=1 (ablation) | SimCLR · CIFAR-10 | test classification accuracy: 80.7% -> 82.1% | Helped |
| Use t-distribution matching and Euclidean feature space | SimCLR · CIFAR-10, 1000 epochs | nearest-neighbor accuracy: 87.2% -> 88.8% | Helped |
| Remove spherical normalization and use negative L2 similarity (ablation) | SimCLR · Gaussian mixture with mean shift | linear classification accuracy: 48.4% -> 100% | Helped |
| Use t-SNE-style matching and Euclidean feature space | MoCo-v2 · DTD domain transfer | accuracy: 71.22 -> 69.20 | Hurt |
Closest papers by meaning
- From $t$-SNE to UMAP with contrastive learning (ICLR 2023 · similarity 0.92)
- Revisiting Contrastive Learning through the Lens of Neighborhood Component Analysis: an Integrated Framework (ICML 2022 · similarity 0.90)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.88)
- Unsupervised visualization of image datasets using contrastive learning (ICLR 2023 · similarity 0.87)
- On the Alignment Between Supervised and Self-Supervised Contrastive Learning (ICLR 2026 · similarity 0.87)
- Single-View Graph Contrastive Learning with Soft Neighborhood Awareness (AAAI 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.