Research map / Label-efficient and robust learning / Self-supervised and contrastive
MetaMask: Revisiting Dimensional Confounder for Self-Supervised Learning
Jiangmeng Li, Wenwen Qiang, Yanan Zhang, Wenyi Mo, Changwen Zheng, Bing Su, Hui Xiong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 MetaMask to CMC | CMC · CIFAR-10, conv | top-1 accuracy: 81.31 -> 85.24 | Improved |
| Add MetaMask to Barlow Twins | Barlow Twins · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 76.63 -> 85.24; 78.49 -> 86.90; 52.80 -> 58.02; 52.95 -> 56.78; 80.97 -> 84.96; 81.43 -> 86.13; 39.81 -> 42.91; 40.34 -> | Improved |
| Add MetaMask to SwAV | SwAV · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 72.19 -> 80.37; 72.36 -> 77.22; 46.58 -> 52.07; 46.75 -> 52.98; 74.65 -> 81.34; 75.30 -> 82.19; 39.67 -> 40.28; 39.02 -> | Improved |
| Add MetaMask to SimCLR | SimCLR · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 80.58 -> 82.85; 80.07 -> 81.64; 50.03 -> 53.70; 49.82 -> 54.24; 75.57 -> 81.37; 77.15 -> 82.90; 36.24 -> 40.56; 39.83 -> | Improved |
| Add MetaMask to BYOL | BYOL · ResNet-18 CIFAR-10 | top-1 accuracy: 88.05 -> 87.53 | Hurt |
| Add MetaMask to BYOL | BYOL · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 77.18 -> 79.26; 80.01 -> 82.73; 53.64 -> 55.68; 53.78 -> 54.72; 41.59 -> 42.72; 41.90 -> 43.29; 81.73 -> 83.01; 81.57 -> | Improved |
| Add MetaMask to SimSiam | SimSiam · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 78.14 -> 80.03; 81.13 -> 81.39; 52.55 -> 54.70; 53.52 -> 55.02; 41.03 -> 41.82; 41.27 -> 42.67; 80.91 -> 82.35; 81.88 -> | Improved |
| Add MetaMask to DCL | DCL · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 80.89 -> 81.33; 80.93 -> 81.34; 51.38 -> 53.05; 51.09 -> 53.74; 38.79 -> 40.28; 40.26 -> 41.69; 77.09 -> 80.94; 78.39 -> | Improved |
| Add MetaMask to HCL | HCL · CIFAR-10, CIFAR-100, STL-10, and IN-200; conv and fc | top-1 accuracy: 82.13 -> 83.65; 82.76 -> 83.95; 52.69 -> 54.03; 53.13 -> 54.26; 40.05 -> 40.43; 41.23 -> 41.01; 79.86 -> 81.52; 80.20 -> | Mixed |
| Add MetaMask to Barlow Twins | Barlow Twins · CIFAR-10, CIFAR-100, STL-10, and IN-200; ResNet-18 | top-1 accuracy: 85.72 -> 87.53; 60.31 -> 61.42; 73.62 -> 73.97; 45.10 -> 46.81 | Improved |
| Add MetaMask to SimCLR | SimCLR · CIFAR-10, CIFAR-100, STL-10, and IN-200; ResNet-18 | top-1 accuracy: 81.73 -> 86.01; 58.27 -> 61.03; 70.09 -> 74.90; 44.46 -> 45.87 | Improved |
| Add MetaMask to NNCLR | NNCLR · CIFAR-10, CIFAR-100, STL-10, and IN-200; ResNet-18 | top-1 accuracy: 81.44 -> 85.78; 61.64 -> 61.89; 71.89 -> 74.02; 47.10 -> 46.90 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Barlow Twins — Barlow Twins: Self-Supervised Learning via Redundancy Reduction (ICML 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Collapse-Proof Non-Contrastive Self-Supervised Learning (ICML 2025 · similarity 0.87)
- Divide and Contrast: Self-Supervised Learning From Uncurated Data (ICCV 2021 · similarity 0.86)
- On the Alignment Between Supervised and Self-Supervised Contrastive Learning (ICLR 2026 · similarity 0.86)
- A theoretical study of inductive biases in contrastive learning (ICLR 2023 · similarity 0.86)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.85)
- Self-Supervised Representation Learning with Meta Comprehensive Regularization (AAAI 2024 · 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.
