Aleksandra Ciprijanovic (Fermilab)
Domain adaptation for cross-domain studies in astronomy
In physics and astronomy, neural networks are often trained on simulated data (source domain) with the prospect of being applied to real detector or telescope images (target domain). Unfortunately, simply training a deep neural network on images from one domain does not guarantee satisfactory performance on new images from a different domain. The ability to share cross-domain knowledge is the main advantage of modern deep domain adaptation techniques. The talk will cover the use of two techniques — Maximum Mean Discrepancy (MMD) and adversarial training with Domain Adversarial Neural Networks (DANN) — for the classification of distant merging galaxies from the Illustris-1 cosmological simulation. The inclusion of domain adaptation methods greatly improves the performance of the algorithm in the target domain when compared to conventional machine learning methods, thereby demonstrating great promise for their use in astronomy.