Research map / Label-efficient and robust learning / Recognition and long-tail
Identifiable Contrastive Learning with Automatic Feature Importance Discovery
Qi Zhang, Yifei Wang, Yisen Wang
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 learnable diagonal importance matrix S (tri-factor) to spectral contrastive loss | SCL · synthetic matrix factorization | distance between optimal solutions: 101.7891 -> 0 | Improved |
| add importance matrix to SCL | SCL ResNet-18 · CIFAR-100 finetune | accuracy: 66.1 -> 67.8 | Improved |
| add importance matrix to SimCLR | SimCLR ResNet-18 · CIFAR-100 finetune | accuracy: 72.5 -> 73.3 | Improved |
| add importance matrix to SCL | SCL ResNet-18 · CIFAR-10 linear | accuracy: 88.4 -> 88.3 | No clear effect |
| add importance matrix to SCL | SCL ResNet-18 · ImageNet-100 linear | accuracy: 72.3 -> 71.2 | Hurt |
| add importance matrix to SCL | SCL ResNet-18 · CIFAR-100 linear | accuracy: 60.3 -> 60.8 | Improved |
| add importance matrix to SimCLR | SimCLR ResNet-18 · ImageNet-100 linear | accuracy: 73.5 -> 74.1 | Improved |
| add importance matrix to SimCLR | SimCLR ResNet-18 · CIFAR-100 linear | accuracy: 60.2 -> 59.8 | No clear effect |
| add importance matrix to SCL | SCL ResNet-18 · CIFAR-10 finetune | accuracy: 92.3 -> 92.7 | Improved |
| add importance matrix to SCL | SCL ResNet-18 · ImageNet-100 finetune | accuracy: 76.4 -> 76.3 | No clear effect |
| add importance matrix to SimCLR | SimCLR ResNet-18 · ImageNet-100 finetune | accuracy: 75.9 -> 75.6 | No clear effect |
| add importance matrix to SimCLR | SimCLR ResNet-18 · CIFAR-10 linear | accuracy: 87.9 -> 87.9 | No clear effect |
Closest papers by meaning
- Revisiting Contrastive Learning through the Lens of Neighborhood Component Analysis: an Integrated Framework (ICML 2022 · similarity 0.87)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.86)
- Non-negative Contrastive Learning (ICLR 2024 · similarity 0.86)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.85)
- On the duality between contrastive and non-contrastive self-supervised learning (ICLR 2023 · similarity 0.85)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.84)
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.
