Work

Application of equivariant transformations in neural networks to classification of galaxy images

Year: 2025

Type: conference-paper

Abstract: We train a deep neural network with SE(2) equivariant layers on galaxy images grouped in two classes - cigar-shaped and round. We use a steerable basis to reduce the memory and computational load produced by the lifting correlations at different orientations of the images from the training dataset. The trained model performs with 99% accuracy in classifying images from the test dataset.

Source: Journal of Physics Conference Series

DOI: 10.1088/1742-6596/3002/1/012007

Authors: Stoyan Mishev, Radoslav Ginin, Yoana Foteva

Institutions: New Bulgarian University, Bulgarian Academy of Sciences, Varna Free University, Trakia University

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: diamond

OpenAlex ID: W4410153147

FWCI:

Cites: 10

Cited by:

Related to: 10