07/24/2023
By Katia Oussar

The Kennedy College of Sciences, Miner School of Computer and Information Sciences, formally invites you to attend the Master's Thesis defense of Katia Oussar entitled "XAI-Infused Vision Transformer for Galaxy Morphology Classification."

Degree: Master's
Date: Wednesday, July 26, 2023
Time: 4:30 p.m. Eastern Time
Location: Via Zoom

Committee Members:

  • Mohammad Arif UI Alam, Ph.D., University of Massachusetts Lowell
  • Ashish Mahabal, Ph.D., California Institute of Technology
  • Cindy Chen, Ph.D., University of Massachusetts Lowell

Abstract:

Galaxy Morphology Classification stands as a fundamental task in astronomy, involving the categorization of galaxies based on their visual and structural features. These intricate patterns offer valuable insights into their formation, evolution, and interactions with cosmic environments. The current era witnesses a remarkable surge in astronomical data, driven by the continuous progress and future prospects of digital sky surveys. Consequently, the astronomy and astrophysics community seek automation to handle the growing data volume effectively. However, several challenges, such as the complexity of large-scale datasets, subjectivity in classification, ambiguous and overlapping features, and the lack of interpretability and explainability, hinder the true potential of this research domain.

In this thesis, we present a cutting-edge vision transformer-based explainable AI technique for Galaxy Morphology classification. Our approach involves incorporating the SWIN (Sifted-Window) Transformer, a state-of-the-art computer vision model known for its effectiveness in handling galaxy morphology classification. The SWIN Transformer serves as a backbone for computer vision tasks, outperforming even the benchmark model, CNN Resnet-50, in general computer vision tasks, including classification. Taking a step further, we integrate eXplainable AI (XAI) into our deep learning application to provide interpretable and transparent insights into our morphology classification decision-making process.

Through extensive evaluation, our research showcases promising results and opens new avenues for further exploration of vision transformer induced XAI applications in the field of Galaxy Morphology classification. This work has the potential to revolutionize the way galaxies are studied and understood, contributing to significant advancements in astronomy and astrophysics.