Method · Optimization and training recipes
Progressive DD
Data Distillation Can Be Like Vodka: Distilling More Times For Better Quality
Xuxi Chen, Yu Yang, Zhangyang Wang, Baharan Mirzasoleiman
ICLR 2024 · first public 2023-10-10 · arXiv 2310.06982
In one paragraph
Argues a single synthetic subset cannot represent the drastically different training dynamics of early versus late training, and instead synthesizes multiple small synthetic subsets in sequence, each conditioned on the ones before it, training on their cumulative union without extra training time; improves existing distillation methods by up to 4.3 points and is the first to scale distillation to considerably larger synthetic datasets.
Where it sits
- Optimization and training recipes (Orthogonal design choices)
- Setting: Image classification
Abstract (verbatim from arXiv)
Dataset distillation aims to minimize the time and memory needed for training deep networks on large datasets, by creating a small set of synthetic images that has a similar generalization performance to that of the full dataset. However, current dataset distillation techniques fall short, showing a notable performance gap when compared to training on the original data. In this work, we are the first to argue that using just one synthetic subset for distillation will not yield optimal generalization performance. This is because the training dynamics of deep networks drastically change during the training. Hence, multiple synthetic subsets are required to capture the training dynamics at different phases of training. To address this issue, we propose Progressive Dataset Distillation (PDD). PDD synthesizes multiple small sets of synthetic images, each conditioned on the previous sets, and trains the model on the cumulative union of these subsets without requiring additional training time. Our extensive experiments show that PDD can effectively improve the performance of existing dataset distillation methods by up to 4.3%. In addition, our method for the first time enable generating considerably larger synthetic datasets.
BibTeX (generated; prefer the venue's official entry)
@article{chen2023data,
title = {Data Distillation Can Be Like Vodka: Distilling More Times For Better Quality},
author = {Xuxi Chen and Yu Yang and Zhangyang Wang and Baharan Mirzasoleiman},
journal = {ICLR 2024},
year = {2023}
}Nearby in Optimization and training recipes
PTM-ST — Multimodal Dataset Distillation via Phased Teacher Models
Shengbin Guo, Hang Zhao, Senqiao Yang et al. · ICLR 2026notableVision–languagepaper ↗code ↗