Understanding Dataset Distillation via Spectral Filtering
Deyu Bo Songhua Liu Xinchao Wang · 2025
Read Understanding Dataset Distillation via Spectral Filtering by Deyu Bo Songhua Liu Xinchao Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-01212.txt.
Description
Deyu Bo, Songhua Liu, and Xinchao Wang seek a shared view of dataset distillation, a way to compress a large dataset into a small one for faster training. Their 2025 paper sets out a framework they call UniDD, based on spectral filtering.
In their view, each distillation method acts as a filter on the eigenvalues of a feature-feature correlation matrix. It also shapes parts of a feature-label correlation matrix by frequency. Low-frequency methods keep global texture, while high-frequency methods keep local detail. Past methods used one fixed filter, which cannot hold both well. UniDD is meant to guide methods that shift across frequencies as distillation proceeds.
Catalog record
| CREATOR | Deyu Bo Songhua Liu Xinchao Wang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-01212