AI, ML and Friends is a weekly seminar series within the School of Computing on Artificial Intelligence, Machine Learning, and related topics. We are open to attendees and presenters external to the school. Please sign up to the mailing list to receive weekly announcements including zoom details, and email the seminar organiser to schedule a talk.
Upcoming Seminars #
07 July 2022, 15:00 #
Wavefront Estimation for Adaptive Optics with Image to Image Translation #
Speaker: Jeffrey Smith
Abstract: The talk will cover the development and evaluation of a new approach to phase retrieval for observational astronomy. Phase retrieval is required where a terrestrial observatory uses an Adaptive Optics (AO) system to assist astronomers in acquiring sharp high-contrast images of faint and distant objects. We have developed a new approach, based on training a conditional adversarial artificial neural network architecture to predict phase using the wavefront sensor data from a closed-loop AO system. Compared to the state-of-the-art model-based approach in astronomy, our approach is not explicitly limited by modelling assumptions–e.g. independence between terms, such as bandwidth and anisoplanatism–and is conceptually simple and flexible. We shall describe our detailed simulation study under different turbulent conditions, using the retrieved residual phase to obtain the impulse response of a simulated instrument. We use the open-source COMPASS tool for simulation, which is the de facto choice for AO research. On key quality metrics, specifically the Strehl ratio and Halo distribution in our application domain, our approach achieves results better than the model-based baseline at a much lower computational cost. We also apply our approach to phase estimation for real-time control of adaptive optics. We develop a GAN Assisted Open Loop (GAOL) design, and compare that with typical linear control in simulation. Using a single trained network, our control experiment demonstrates significant improvement across a range of typical turbulence conditions.
Bio: Jeffrey Smith is a PhD student in the School of Computing at the Australian National University. Jeff is supervised by Charles Gretton from the ANU school of computing, and Damien Gratadour who is jointly appointed at LESIA Observatoire de Paris and at the ANU Research School of Astronomy and Astrophysics (RSAA). Jeff’s talk will describe work done with Jesse Cranney, also from RSAA. Jeff holds qualifications in optoelectronics and computing at Macquarie University, Australia, including the degrees of Master of Information Technology and Master of Research in Experimental Mathematics. Jeff is a keen apiarist, and has extensive practical experience working in the Sydney financial markets.
Where: Building 145, room 1.33