Research map / Label-efficient and robust learning / Self-supervised and contrastive
$i$-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning
Kibok Lee, Yian Zhu, Kihyuk Sohn, Chun-Liang Li, Jinwoo Shin, Honglak Lee
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 i-Mix to N-pair | N-pair contrastive learning · CIFAR-10 | accuracy: 93.3 -> 95.6 | Improved |
| Add i-Mix to MoCo v2 | MoCo v2 · CIFAR-100 | accuracy: 71.6 -> 78.1 | Improved |
| Add i-Mix to BYOL | BYOL · Speech Commands | accuracy: 94.8 -> 98.3 | Improved |
| Add i-Mix to MoCo v2 | MoCo v2 · CovType | accuracy: 70.5 -> 73.1 | Improved |
| Add i-Mix with no other augmentations (ablation) | MoCo v2 · CIFAR-10 | accuracy: 47.7 -> 83.4 | Improved |
| Add i-Mix with no other augmentations (ablation) | MoCo v2 · Speech Commands | accuracy: 76.9 -> 92.8 | Improved |
| Add i-Mix with no other augmentations (ablation) | MoCo v2 · Higgs1M | accuracy: 65.5 -> 72.9 | Improved |
| Add InputMix alongside i-Mix (ablation) | i-Mix on CIFAR-10 without other data augmentations · CIFAR-10 | accuracy: 75.1% to 83.4% | Improved |
| Use i-CutMix instead of i-MixUp (ablation) | N-pair and SimCLR · CIFAR-100 pretext to CIFAR-10 downstream | accuracy: 86.2 vs. 85.1; 85.2 vs. 83.7 | Mixed |
| Use i-CutMix instead of i-MixUp (ablation) | N-pair and SimCLR · CIFAR-10 pretext to CIFAR-100 downstream | accuracy: 63.3 vs. 61.5; 61.4 vs. 57.1 | Mixed |
| Apply i-Mix to supervised contrastive learning (ablation) | N-pair supervised contrastive learning · CIFAR-10 | accuracy: 95.7% to 97.0% | Improved |
| Apply i-Mix to supervised contrastive learning with CIFAR-10 pretext (ablation) | N-pair supervised contrastive learning · CIFAR-100 downstream | accuracy: 58.9 to 57.8 | Hurt |
Closest papers by meaning
- Towards Domain-Agnostic Contrastive Learning (ICML 2021 · similarity 0.89)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.88)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.88)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.87)
- TiMix: Text-Aware Image Mixing for Effective Vision-Language Pre-training (AAAI 2024 · similarity 0.87)
- DiffAug: Enhance Unsupervised Contrastive Learning with Domain-Knowledge-Free Diffusion-based Data Augmentation (ICML 2024 · similarity 0.87)
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.
