New method uses diffusion models for efficient dataset distillation

Researchers have developed a novel framework for dataset distillation that leverages pre-trained diffusion models for patch selection rather than direct image generation. This method addresses challenges like distribution shifts and multi-step distillation by predicting noise from the diffusion model to identify distinctive image regions. The approach then applies intra-class clustering and ranking to ensure patch diversity, enabling a streamlined, one-step distillation process. Experiments show this method consistently outperforms existing techniques on metrics and settings for large-scale datasets like ImageNet-1K with complex networks such as ResNet-101. AI

IMPACT This research could lead to more efficient training of deep learning models by reducing the need for large datasets.

RANK_REASON The cluster contains an academic paper detailing a new method in computer vision. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• ResNet-101
• Xinhao Zhong

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →