avatar
AstroAI
Developing Artificial Intelligence to Solve the Mysteries of the Universe
  • HOME
  • RESEARCH
  • EARTHAI
  • PEOPLE
  • EVENTS
  • LATEST NEWS
  • LUNCH TALKS
  • SUMMER PROGRAM
  • WORKSHOP
  • APPLY
  • CONTACT
Home AstroAI Workshop 2025
Workshop_abstract
Cancel

AstroAI Workshop 2025

Details Invited Speakers Abstracts Register Schedule Venue Accommodations Code of Conduct

Srinadh Reddy Bhavanam

MargFormer: Photometric Classification of Stars, Quasars and Compact Galaxies with Cross-Attention Vision Transformer

Presenter: Srinadh Reddy Bhavanam

Title: MargFormer: Photometric Classification of Stars, Quasars and Compact Galaxies with Cross-Attention Vision Transformer

Date/Time: Monday, July 7th, 3:30 - 5:00 PM

Abstract: We present MargFormer, a unified deep-learning model designed to classify stars, quasars, and compact galaxies by synergistically integrating photometric parameters and imaging data. Leveraging a cross-attention vision transformer where photometric features serve as queries to probe imaging data, MargFormer processes both heterogeneous data types within a single cohesive framework. This unified architecture contrasts with conventional approaches, where prior methods process each data modality of photometric parameters and images using corresponding architectures such as ANNs and CNNs and then stack their outputs. By enabling photometry to guide image feature extraction within this joint framework, MargFormer effectively captures intricate local/global features and cross-modal correlations. This results in a substantially lightweight model with fewer parameters and demonstrates improved generalization performance. Evaluated on data from the Sloan Digital Sky Survey (SDSS) Data Release (DR) 16, MargFormer achieves performance comparable to or exceeding state-of-the-art methods, even at fainter magnitudes. This work underscores the power and efficiency of transformer-based models, highlighting their potential as scalable solutions with strong generalization capabilities, crucial for analyzing the vast and diverse datasets from upcoming wide-field surveys like the Vera C. Rubin Observatory. Our trained models and code are publicly available at https://github.com/srinadh99/AstroFormer/tree/main.

-->

© 2025 AstroAI. Some rights reserved.

Powered by Jekyll with Chirpy theme.

A new version of content is available.