Image Distillation with Machine Learning Image Modification and K-Means Clustering

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Mathematics

Date of Award

Summer 8-18-2025

Abstract

Image dataset distillation synthesizes a representative set which preserves the features of the original larger training set that leads to significant decreasing of computer resources. This work develops efficient dataset distillation method based on gradient image modification with k-means clustering representation for model training. It targets significant reduction in dataset size while maintaining model performance. First, we modify the training images, then apply k-means method to cluster them, after that we take the mean of every cluster as distilled images. Experimental results for the proposed approach are conducted on the benchmark image datasets Digit-MNIST, Fashion-MNIST, and CIFAR-10. The thesis completes with comparing our experimental result with those obtained by contemporary method.

Advisor

Nikolay Sirakov

Subject Categories

Mathematics | Physical Sciences and Mathematics

Share

COinS