Though machine learning has been used to great effect for decades in astronomy to accelerate discovery, generative modeling with neural architectures has been used in only limited contexts. In this talk, I focus on some of the new practical implementations and uses for generative models in astronomy. One application arises in the need to optimize telescope observing cadences, requiring the generation of physically plausible astronomical time-series. I present our approach to this using semi-supervised variational autoencoders where physical inputs are mapped to the (generative) latent space. I also present our recent work on a successful fast imaging artifact (cosmic rays) discovery and inpainting framework; improving the efficiency for how future data can be taking, this presents clear cost implications for major upcoming space and ground base missions. Last, I will comment upon the potential use of semantic indexing with autoencoders for compressed astronomical sensing.
Back to Workshop I: From Passive to Active: Generative and Reinforcement Learning with Physics