« search calendars« DIMACS Workshop on ADMM and Proximal Splitting Methods in Optimization

« Source Separation in Astronomy with Constrained Matrix Factorization

Source Separation in Astronomy with Constrained Matrix Factorization

June 13, 2018, 2:30 PM - 3:00 PM



Rutgers University

CoRE Building

96 Frelinghuysen Road

Piscataway, NJ 08854

Click here for map.

Peter Melchior, Princeton University

Modern astronomical images are collected with increasingly large telescopes and reach spectacular sensitivities. As an undesired consequence, images have become so densely populated with stars and galaxies that those sources routinely overlap, which interferes with the measurement process.

We employ a matrix factorization scheme to perform source separation across multiple images, observed in different optical filters. We use a combination of proximal gradient and ADMM approaches to enforce multiple simultaneous constraints on the spectral energy emission and/or morphology of the sources. I will present the methodology and show results of the source separation framework applied to state-of-the-art astronomical images from the HyperSuprimeCam Survey based in Hawaii.


Slides     Video