Research map / Label-efficient and robust learning / Self-supervised and contrastive
Revisiting Contrastive Learning through the Lens of Neighborhood Component Analysis: an Integrated Framework
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 |
|---|---|---|---|
| Increase the neighborhood positive count M above 1 (ablation) | NCA / NaCl · CIFAR100 | accuracy: 53.69±0.25 -> 57.32±0.17 | Helped |
| Increase the neighborhood positive count M above 1 (ablation) | NCA / NaCl · CIFAR100 | FGSM accuracy: 25.17±0.55 -> 28.41±0.24 | Helped |
| Use MIXNCA with λ=0.5 rather than the M=1 loss (ablation) | NaCl · CIFAR100 | FGSM accuracy: 25.17±0.55 -> 24.01±0.91 | Hurt |
| Add the robust loss with adversarial weighting (ablation) | IntCl (M=1, g2) · CIFAR100 | FGSM accuracy: 31.03% to 40.05% | Helped |
| Add the robust loss (ablation) | NCA (g2, M=3) · CIFAR100 | FGSM accuracy: 33.19% to 40.53% | Helped |
| Add the robust loss with adversarial weighting (ablation) | MIXNCA (g2, M=2, λ=0.5) · TinyImageNet | FGSM accuracy: 28.29% to 32.65% | Helped |
| Increase M from 1 to 5 (ablation) | IntNaCl MIXNCA (g2, λ=0.5) · CIFAR100 | accuracy: 56.22% to 59.43% | Helped |
| Increase M from 1 to 5 (ablation) | IntNaCl MIXNCA (g2, λ=0.9) · CIFAR100 | accuracy: 56.22% to 56.86% | Helped |
| Use goal-specific adversarial weighting w(x)=ŵ(x) rather than w(x)=1 (ablation) | IntNaCl MIXNCA (g2, M=5, λ=0.5) · TinyImageNet | accuracy: 44.77±0.11 -> 45.31±0.22 | Helped |
| Use goal-specific adversarial weighting w(x)=ŵ(x) rather than w(x)=1 (ablation) | IntNaCl MIXNCA (g2, M=5, λ=0.5) · TinyImageNet | FGSM accuracy: 32.47±0.42 -> 32.43±0.33 | No clear effect |
Closest papers by meaning
- Your Contrastive Learning Is Secretly Doing Stochastic Neighbor Embedding (ICLR 2023 · similarity 0.90)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.88)
- Single-View Graph Contrastive Learning with Soft Neighborhood Awareness (AAAI 2025 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.87)
- Contrastive Learning is Spectral Clustering on Similarity Graph (ICLR 2024 · similarity 0.87)
- Non-negative Contrastive Learning (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.