Machine learning (ML) has revolutionized numerous scientific domains in the last decade. In particular, deep neural networks are surprisingly good at learning the inherent patterns in various sorts of data and mapping them to specific outcomes of our interest. The learned representations generalize well to the unseen dataset, making ML a powerful algorithm for making predictions with limited information or data that have complicated structures. In this talk, I will introduce my research on leveraging ML to (i) study strong gravitational lens systems observed by big telescopes; and (ii) help discover neutrinos from core-collapse supernovae that have ever happened throughout the cosmic history in the Super-Kamiokande. Our results show that ML is highly efficient in capturing the underlying physics and extracting the hidden features in the datasets of both problems, thus opening exciting prospects in astronomy and particle physics.
