Project:	Discovering rare galaxies in survey imagery with self-supervised anomaly detection (JPL AO No. 16823)
Disciplines:	Astronomy, Data Science
Mentor:	Eric Huff, (JPL), Eric.M.Huff@jpl.nasa.gov, Phone: (626) 460-9834
Background:	Modern galaxy surveys observe billions of galaxies across multiple wavelength bands, effectively capturing different “colors” of light from visible to near-infrared. For the majority of galaxies, these bands follow strong, predictable relationships driven by stellar populations. Galaxies that deviate from these relationships in unexpected ways are likely to be of scientific interest, including gravitationally lensed galaxies, extremely distant galaxies, unusual mergers, and other exotic systems. Studying these outlier galaxies provides a way to test of our models of galaxy evolution, dark matter and gravity in the most extreme physical cases. Given the sheer number of galaxies in modern galaxy surveys, identifying unusual galaxies manually is impractical. Self-supervised learning offers an efficient and scalable way to identify rare and unique objects, without human intervention and labeling.
Description:	The student will design and develop a self-supervised deep learning model to detect unusual objects in multi-band galaxy images from the early data release of the Euclid Space Telescope. The core task will involve training a vision model to learn typical cross-band relationships by predicting one band from the others, and pairing this with outlier detection techniques to identify interesting objects that exhibit unpredictable patterns. The student will be responsible for implementing the model, evaluating its performance, and visualizing and interpreting the detected anomalies. Expected outcomes include a catalog of anomalous galaxy candidates and a characterization of their properties through image inspection and visualization. The resulting pipeline can in future be scaled to anomaly detection in the full Euclid dataset.
Student Requirements:	Required: - Familiarity with Python - Experience implementing and training basic deep learning models in your deep learning framework of choice, such as PyTorch, TensorFlow, or Jax. - Introductory knowledge of astronomy and galaxies - General statistics and linear algebra skills
Location / Safety:	Project building and/or room locations: . Student will need special safety training: .
Programs:	This AO can be done under the following programs:   Program    Available To        SURF@JPL    both Caltech and non-Caltech students  Click on a program name for program info and application requirements.