Research map / Label-efficient and robust learning / Recognition and long-tail
Diversity-Enhanced Distribution Alignment for Dataset Distillation
Hongcheng Li, Yucan Zhou, Xiaoyan Gu, Bo Li, Weiping 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 class-specific mean alignment from stored Gaussian statistics in place of BN alignment (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy: 47.5 -> 63.1 | Improved |
| add covariance alignment to mean alignment (ablation) | DEDA · CIFAR-100 IPC=10 | top-1 accuracy: 39.8 -> 41.6 | Improved |
| add covariance alignment to mean alignment (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy: 63.1 -> 63.8 | Improved |
| add last-layer variance maximization regularizer LVar (ablation) | DEDA · CIFAR-100 | top-1 accuracy: +0.6% and 0.9% | Improved |
| add off-diagonal covariance minimization LCov (ablation) | DEDA · CIFAR-100 | top-1 accuracy: +1.4% and 0.4% | Improved |
| add LCov without LVar (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy: 63.8 -> 64.2 | Improved |
| increase covariance weight gamma from 0 to 50 (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy: 0 -> 50 | Improved |
| increase gamma beyond 50 to 200 (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy: 50 -> 200 | Hurt |
| use large lambda1 for variance regularization (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy | Hurt |
| vary lambda2 covariance regularization weight (ablation) | DEDA · CIFAR-100 IPC=50 | top-1 accuracy | No clear effect |
| replace SRe2L BN alignment with DEDA, ResNet-18 distillation | DEDA vs SRe2L · CIFAR-10 IPC=50 | top-1 accuracy: 47.5 -> 60.2 | Improved |
| replace SRe2L with DEDA on ResNet-50 cross-architecture | DEDA vs SRe2L · CIFAR-100 IPC=50 cross-arch | top-1 accuracy: 52.8 -> 63.7 | Improved |
Closest papers by meaning
- Exploiting Inter-sample and Inter-feature Relations in Dataset Distillation (CVPR 2024 · similarity 0.89)
- Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation (NeurIPS 2025 · similarity 0.89)
- DELT: A Simple Diversity-driven EarlyLate Training for Dataset Distillation (CVPR 2025 · similarity 0.88)
- Provable and Efficient Dataset Distillation for Kernel Ridge Regression (NeurIPS 2024 · similarity 0.87)
- Large Scale Dataset Distillation with Domain Shift (ICML 2024 · similarity 0.87)
- Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation (NeurIPS 2025 · 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.
