Research map / Label-efficient and robust learning / Recognition and long-tail
Theoretical Analysis of Contrastive Learning under Imbalanced Data: From Training Dynamics to a Pruning Solution
Haixu Liao, Yating Zhou, Songyang Zhang, Meng Wang, Shuai Zhang
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 magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=10) | linear probe accuracy (%): 79.25 ± 1.03 -> 84.92 ± 0.67 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=50) | linear probe accuracy (%): 75.58 ± 0.84 -> 83.60 ± 1.02 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=100) | linear probe accuracy (%): 74.24 ± 0.82 -> 81.31 ± 0.94 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR100-LT (rho=10) | linear probe accuracy (%): 51.21± 1.21 -> 56.33±1.51 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR100-LT (rho=50) | linear probe accuracy (%): 49.32± 0.45 -> 56.12± 0.32 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR100-LT (rho=100) | linear probe accuracy (%): 47.12±0.51 -> 54.93±0.50 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · ImageNet-LT (rho=256) | linear probe accuracy (%): 63.21 -> 65.12 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=10) | head-tail accuracy gap Delta20: 3.42±1.02 -> 2.99±0.92 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=50) | head-tail accuracy gap Delta20: 3.92±1.21 -> 3.35± 0.76 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR10-LT (rho=100) | head-tail accuracy gap Delta20: 5.69±1.35 -> 5.62±0.99 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR100-LT (rho=10) | head-tail accuracy gap Delta20: 2.45±0.57 -> 1.37±0.46 | Improved |
| add magnitude-based neuron pruning to contrastive pretraining | contrastive learning · CIFAR100-LT (rho=50) | head-tail accuracy gap Delta20: 4.95±1.02 -> 2.57±0.92 | Improved |
Closest papers by meaning
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.87)
- An Investigation of Representation and Allocation Harms in Contrastive Learning (ICLR 2024 · similarity 0.86)
- BaCon: Boosting Imbalanced Semi-supervised Learning via Balanced Feature-Level Contrastive Learning (AAAI 2024 · similarity 0.86)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.86)
- Symmetric Neural-Collapse Representations with Supervised Contrastive Loss: The Impact of ReLU and Batching (ICLR 2024 · similarity 0.86)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 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.
