Saved in:
Bibliographic Details
Main Authors: Yang, William, Zhu, Ye, Deng, Zhiwei, Russakovsky, Olga
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.04284
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910536438906880
author Yang, William
Zhu, Ye
Deng, Zhiwei
Russakovsky, Olga
author_facet Yang, William
Zhu, Ye
Deng, Zhiwei
Russakovsky, Olga
contents Dataset distillation has emerged as a strategy to overcome the hurdles associated with large datasets by learning a compact set of synthetic data that retains essential information from the original dataset. While distilled data can be used to train high performing models, little is understood about how the information is stored. In this study, we posit and answer three questions about the behavior, representativeness, and point-wise information content of distilled data. We reveal distilled data cannot serve as a substitute for real data during training outside the standard evaluation setting for dataset distillation. Additionally, the distillation process retains high task performance by compressing information related to the early training dynamics of real models. Finally, we provide an framework for interpreting distilled data and reveal that individual distilled data points contain meaningful semantic information. This investigation sheds light on the intricate nature of distilled data, providing a better understanding on how they can be effectively utilized.
format Preprint
id arxiv_https___arxiv_org_abs_2406_04284
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle What is Dataset Distillation Learning?
Yang, William
Zhu, Ye
Deng, Zhiwei
Russakovsky, Olga
Machine Learning
Dataset distillation has emerged as a strategy to overcome the hurdles associated with large datasets by learning a compact set of synthetic data that retains essential information from the original dataset. While distilled data can be used to train high performing models, little is understood about how the information is stored. In this study, we posit and answer three questions about the behavior, representativeness, and point-wise information content of distilled data. We reveal distilled data cannot serve as a substitute for real data during training outside the standard evaluation setting for dataset distillation. Additionally, the distillation process retains high task performance by compressing information related to the early training dynamics of real models. Finally, we provide an framework for interpreting distilled data and reveal that individual distilled data points contain meaningful semantic information. This investigation sheds light on the intricate nature of distilled data, providing a better understanding on how they can be effectively utilized.
title What is Dataset Distillation Learning?
topic Machine Learning
url https://arxiv.org/abs/2406.04284