Research map / Label-efficient and robust learning / Self-supervised and contrastive
Combating Representation Learning Disparity with Geometric Harmonization
Zhihan Zhou, Jiangchao Yao, Feng Hong, Ya Zhang, Bo Han, Yanfeng 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 GH to SimCLR | SimCLR · CIFAR-100-LT IR100 | linear probing Avg: 50.72 -> 53.96 | Improved |
| add GH to BCL | BCL · CIFAR-100-LT IR100 | linear probing Avg: 56.45 -> 57.65 | Improved |
| add GH to baselines | SimCLR/Focal/SDCLR/DnC/BCL · Places-LT | Std of group accuracy: 1.83 -> 1.89 | No clear effect |
| add GH to baselines | SimCLR · ImageNet-LT | Std of group accuracy: 5.19 -> 3.15 | Improved |
| remove geometric uniform structure (GUS) (ablation) | GH · CIFAR-100-LT IR100 | linear probing: 53.3 -> 54.0 | Improved |
| remove class prior pi (ablation) | GH · CIFAR-100-LT IR100 | linear probing: 53.1 -> 54.0 | Improved |
| replace GH with SwAV clustering (ablation) | SimCLR · CIFAR-100-LT IR100 | linear probing: 54.0 -> 52.2 | Hurt |
| replace GH with SeLA clustering (ablation) | SimCLR · CIFAR-100-LT IR100 | linear probing: 50.7 -> 50.5 | No clear effect |
| use joint instead of bi-level optimization (ablation) | GH · CIFAR-100-LT IR100 | linear probing: 53.96 -> 50.18 | Hurt |
| add warm-up to joint optimization (ablation) | GH joint · CIFAR-100-LT IR100 | linear probing: 50.18 -> 51.14 | Mixed |
| use GH-pretrained init for logit adjustment | Logit adjustment · Places-LT | accuracy: 32.63 -> 33.86 | Improved |
| add GH to SimCLR pretraining on CC3M | SimCLR · ImageNet finetune | accuracy: 52.06 -> 53.39 | Improved |
Closest papers by meaning
- Self-supervised Learning is More Robust to Dataset Imbalance (ICLR 2022 · similarity 0.86)
- Self-supervised Representation Learning from Random Data Projectors (ICLR 2024 · similarity 0.86)
- Rethinking the Uniformity Metric in Self-Supervised Learning (ICLR 2024 · similarity 0.86)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.85)
- Can Generative Models Improve Self-Supervised Representation Learning? (AAAI 2025 · similarity 0.85)
- Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections (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.
