Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Damaging Contrastive 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 an online-pruned self-competitor branch | SimCLR on CIFAR10-LT · CIFAR10-LT | linear separability, All accuracy: 75.55 ± 0.66 -> 82.00 ± 0.68 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on CIFAR100-LT · CIFAR100-LT | linear separability, All accuracy: 47.11 ± 0.34 -> 55.48 ± 0.62 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on ImageNet-100-LT · ImageNet-100-LT | linear separability, All accuracy: 65.46 -> 66.48 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on CIFAR10 · CIFAR10 few-shot | All accuracy: 67.08 ± 2.15 -> 70.47 ± 1.38 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on CIFAR100 · CIFAR100 few-shot | All accuracy: 22.96 ± 0.43 -> 25.27 ± 0.83 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on ImageNet-100-LT · ImageNet-100-LT few-shot | All accuracy: 42.16 -> 42.38 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on CIFAR10 · CIFAR10 linear evaluation | Accuracy: 91.16 -> 91.55 | Helped |
| Add an online-pruned self-competitor branch | SimCLR on CIFAR100 · CIFAR100 linear evaluation | Accuracy: 62.84 -> 66.32 | Helped |
| Use random dropout instead of magnitude-based pruning to form the sparse branch (ablation) | SDCLR on CIFAR100-LT · CIFAR100-LT | linear separability accuracy: 21.99±0.35% | Hurt |
| Apply focal loss with gamma 2.0 (ablation) | SimCLR on CIFAR100-LT · CIFAR100-LT | linear separability accuracy: 47.33±0.33% -> 46.48±0.51% | Hurt |
| Vary the pruning ratio for the sparse branch (ablation) | SDCLR on CIFAR100-LT · CIFAR100-LT | linear separability accuracy | Mixed |
| Update the pruning mask more frequently than once per epoch (ablation) | SDCLR · Training | performance | No clear effect |
Closest papers by meaning
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.90)
- Self-Contrastive Learning: Single-Viewed Supervised Contrastive Framework Using Sub-network (AAAI 2023 · similarity 0.89)
- Exploring Balanced Feature Spaces for Representation Learning (ICLR 2021 · similarity 0.89)
- A theoretical study of inductive biases in contrastive learning (ICLR 2023 · similarity 0.89)
- An Investigation of Representation and Allocation Harms in Contrastive Learning (ICLR 2024 · similarity 0.88)
- BECLR: Batch Enhanced Contrastive Few-Shot Learning (ICLR 2024 · similarity 0.88)
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.