Frequency Domain-Based Dataset Distillation

Authors: Donghyeok Shin, Seungjae Shin, Il-chul Moon

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through the selection of frequency dimensions based on the explained variance, Fre D demonstrates both theoretical and empirical evidence of its ability to operate efficiently within a limited budget, while better preserving the information of the original dataset compared to conventional parameterization methods. Furthermore, based on the orthogonal compatibility of Fre D with existing methods, we confirm that Fre D consistently improves the performances of existing distillation methods over the evaluation scenarios with different benchmark datasets. ... We evaluate the efficacy of Fre D on various benchmark datasets, i.e. SVHN [13], CIFAR-10, CIFAR-100 [8] and Image Net-Subset [6, 3, 4].
Researcher Affiliation Collaboration Donghyeok Shin KAIST tlsehdgur0@kaist.ac.kr Seungjae Shin KAIST tmdwo0910@kaist.ac.kr Il-Chul Moon KAIST, Summary.AI icmoon@kaist.ac.kr
Pseudocode Yes Algorithm 1 Fre D: Frequency domain-based Dataset Distillation
Open Source Code Yes We release the code at https://github.com/sdh0818/Fre D.
Open Datasets Yes We evaluate the efficacy of Fre D on various benchmark datasets, i.e. SVHN [13], CIFAR-10, CIFAR-100 [8] and Image Net-Subset [6, 3, 4].
Dataset Splits No The paper mentions training and evaluation but does not explicitly detail training/validation/test splits (e.g., specific percentages, sample counts, or explicit mention of a validation set beyond general deep learning context).
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., 'Python 3.8, PyTorch 1.9, and CUDA 11.1').
Experiment Setup Yes We evaluate each method by training 5 randomly initialized networks from scratch on optimized S. Please refer to Appendix C for a detailed explanation of datasets and experiment settings.