How time-domain astronomers use ML on streaming, noisy, distorted images of the sky — the search for Planet 9, discovery of new variable sources, data-driven emulators — and building physics-informed learning architectures.