Research map / Label-efficient and robust learning / Self-supervised and contrastive
MaskCon: Masked Contrastive Learning for Coarse-Labelled Dataset
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 masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · CIFARtoy-goodsplit | Recall@1: 86.15 -> 90.28 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · CIFARtoy-badsplit | Recall@1: 90.55 -> 91.56 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · CIFAR100 | Recall@1: 60.57 -> 65.52 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · ImageNet-1K (32×32) | Recall@1: 18.36 -> 19.08 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · SOP-split1 | Recall@1: 74.02 -> 74.05 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · SOP-split2 | Recall@1: 39.12 -> 45.36 | Helped |
| Use masked, similarity-weighted coarse-class relations instead of the strongest coarse-label baseline | MaskCon · Stanford Cars196 | Recall@1: 42.77 -> 45.53 | Helped |
| Include additional inter-sample relations (w≠0) rather than self-supervised contrastive learning alone (ablation) | MaskCon · CIFAR100 | Recall@1: 40.50 -> 65.52 | Helped |
| Use a suitable finite temperature (τ=0.05) rather than τ=∞, which reduces to Grafit (ablation) | MaskCon · CIFAR100 | Recall@1: 60.57 -> 65.52 | Helped |
| Set τ=0.01 instead of τ=0.05 (ablation) | MaskCon (w=1) · CIFAR100 | Recall@1: 65.52 -> 61.41 | Hurt |
| Set τ=0.1 instead of τ=0.05 (ablation) | MaskCon (w=1) · CIFAR100 | Recall@1: 65.52 -> 62.52 | Hurt |
| Set τ=0.5 instead of τ=0.05 (ablation) | MaskCon (w=1) · CIFAR100 | Recall@1: 65.52 -> 59.76 | Hurt |
Closest papers by meaning
- Layer Grafted Pre-training: Bridging Contrastive Learning And Masked Image Modeling For Label-Efficient Representations (ICLR 2023 · similarity 0.88)
- Contrastive Tuning: A Little Help to Make Masked Autoencoders Forget (AAAI 2024 · similarity 0.87)
- Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners (ICLR 2024 · similarity 0.86)
- Efficient Vision-Language Pre-training by Cluster Masking (CVPR 2024 · similarity 0.86)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.86)
- Multi-Modal Contrastive Masked Autoencoders: A Two-Stage Progressive Pre-training Approach for RGBD Datasets (CVPR 2025 · similarity 0.85)
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.