26th Australian Institute of Physics Congress and the Australian and New Zealand Conference on Optics and Photonics

Differentiable Simulation in Astronomy and Optics

Not scheduled

20m

Belinda Hutchinson Building (The University of Sydney )

Contributed Oral AIP | Astronomy (ASTRO)

Description

Deep learning is underpinned by software allowing efficient calculation of exact derivatives. Writing physics simulations in frameworks for this, like Jax or PyTorch, allows for revolutionary new capabilities: optimization and inference in extremely high dimensions. With our dLux Fresnel-optics code, we can fit optical models jointly with neural networks to the James Webb and Hubble Space Telescopes; to design the Toliman Space Telescope; and to learn the fundamental information content of optical imaging interferometry. I will talk about these projects and situate them in the general context of an emerging ecosystem of differentiable simulation in physics and astronomy.