I recently returned to astronomy after careers in software development and politics. My work involves the application of modern machine learning techniques, particularly deep neural networks, to astronomical discovery. This involved training computers to find rare gravitational lenses among hundreds of millions of sources in imaging surveys. Current work involves teaching machines to understand galaxy spectral data and the astrophysics that underlie them. I am a postdoc working on Prof. Karl Glazebrook's Laureate Fellowship, focused on understanding the earliest galaxies that will be revealed by the James Webb Space Telescope.
Probing neural networks for science: What is it they are learning?
Neural Networks are finding increasing use in many areas of astronomy, but often act as "black boxes". Many techniques exist to probe in the internals of neural networks but not all are relevant to scientists. In this talk I discuss some of the techniques developed in computer vision to investigate what neural networks are learning, and investigate some of their benefits and problems when applied to astronomy. I introduce a simple technique and software package, 'Sensie', to probe what neural networks have learned. I apply it to networks trained to find strong gravitational lenses in the Dark Energy Survey and draw some lessons that may help future searches.