Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Contrastive Learning: Single-Viewed Supervised Contrastive Framework Using Sub-network
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 |
|---|---|---|---|
| replace augmentation-based views with sub-network feature views in a single-view supervised contrastive framework | SelfCon · CIFAR-100 | top-1 accuracy: 73.9 -> 75.4 | Helped |
| replace augmentation-based views with sub-network feature views in a single-view supervised contrastive framework | SelfCon · ImageNet | accuracy: 78.0^{\ddag} -> 78.6 | Helped |
| use a single-view batch instead of a multi-view batch (ablation) | SelfCon · ImageNet, ResNet-18 | accuracy: 71.6 -> 71.4 | Hurt |
| use a single-view batch instead of a multi-view batch (ablation) | SelfCon · CIFAR-100, ResNet-18 | accuracy: 74.9 -> 75.4 | Helped |
| add a sub-network to the single-view contrastive method (ablation) | SupCon-S · ImageNet, ResNet-50 | accuracy: 77.5 -> 78.6 | Helped |
| ensemble backbone and sub-network predictions | SelfCon · ImageNet | accuracy: 78.6 -> 79.5; +3.0 vs. CE; +1.5 vs. SupCon | Helped |
| use an FC sub-network after the 2nd block rather than a Small sub-network (ablation) | SelfCon, ResNet-18 · CIFAR-100 | accuracy: 74.7_{\pm{0.2}} -> 75.4_{\pm{0.1}} | Helped |
| use an FC sub-network after the 2nd block rather than a Same sub-network (ablation) | SelfCon, ResNet-18 · CIFAR-100 | accuracy: 74.5_{\pm{0.0}} -> 75.4_{\pm{0.1}} | Helped |
| move the FC sub-network exit from the 2nd block to the 1st block (ablation) | SelfCon, ResNet-18 · CIFAR-100 | accuracy: 75.4_{\pm{0.1}} -> 73.2_{\pm{0.2}} | Hurt |
| move the FC sub-network exit from the 2nd block to the 3rd block (ablation) | SelfCon, ResNet-18 · CIFAR-100 | accuracy: 75.4_{\pm{0.1}} -> 75.5_{\pm{0.1}} | No clear effect |
| add FC exits after all three blocks instead of using one exit after the 2nd block (ablation) | SelfCon, ResNet-18 · CIFAR-100 | accuracy: 75.4_{\pm{0.1}} -> 74.5_{\pm{0.1}} | Hurt |
| ensemble backbone and sub-network predictions | SelfCon · CIFAR-100 | accuracy: 78.5 -> 80.0; +5.2 vs. CE; +4.5 vs. SupCon | Helped |
Closest papers by meaning
- Supervised Contrastive Learning (NeurIPS 2020 · similarity 0.93)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.92)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.92)
- On the duality between contrastive and non-contrastive self-supervised learning (ICLR 2023 · similarity 0.91)
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.91)
- An Augmentation-Aware Theory for Self-Supervised Contrastive Learning (ICML 2025 · 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.