Research map / Label-efficient and robust learning / Recognition and long-tail
Long-Tailed Recognition by Mutual Information Maximization between Latent Features and Ground-Truth Labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| maximize mutual information between latent features and ground-truth labels rather than input data | GML loss · ImageNet-LT | accuracy (All) | Helped |
| add the teacher-student strategy to GML (ablation) | Proposed method · ImageNet-LT | accuracy: 56.0 -> 58.3 (+2.3) | Helped |
| add the teacher-student strategy (ablation) | BCL · ImageNet-LT | accuracy: 56.7 -> 57.1 (+0.4) | No clear effect |
| use the proposed loss instead of cross-entropy (ablation) | ImageNet-LT training · ImageNet-LT | accuracy: 55.4 -> 56.0 | Helped |
| increase teacher backbone performance from ResNet-34 to ResNeXt-101 (ablation) | Proposed teacher-student training · ImageNet-LT | student accuracy: 58.0 -> 58.3 | Helped |
| train τ_g instead of fixing it (ablation) | Proposed method · ImageNet-LT | accuracy: 58.2 -> 58.3 | Helped |
| use logit-adjustment strength α=1.0 instead of α=0.2 (ablation) | Proposed segmentation loss · ADE20K | mIoU: 38.1 -> 31.7 | Hurt |
| use logit-adjustment strength α=1.0 instead of α=0.2 (ablation) | Proposed segmentation loss · ADE20K | mAcc: 51.4 -> 59.8 | Helped |
| use the proposed method rather than supervised contrastive learning | CIFAR-10-LT · CIFAR-10-LT (imbalance factor 1) | accuracy: 96.0 -> 95.9 | No clear effect |
| use the proposed method rather than supervised contrastive learning | CIFAR-10-LT · CIFAR-10-LT (imbalance factor 100) | accuracy: 82.7 -> 86.7 | Helped |
| replace cross-entropy with the proposed loss | ADE20K segmentation · ADE20K | mIoU: 36.1 -> 38.1 | Helped |
Closest papers by meaning
- Decoupled Contrastive Learning for Long-Tailed Recognition (AAAI 2024 · similarity 0.91)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.91)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.88)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.87)
- DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge (NeurIPS 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.