Date: Tuesday 11 April 2023 - 10:00 BST (Europe/London)
Speaker: Jesse Granney (Postdoctoral Fellow at Australian National University College of Science) and Charles Gretton (TechLauncher Program Convener, at Australian National University College of Engineering, Computing and Cybernetics)
Abstract
Atmospheric turbulence severely limits the quality of science able to be retrieved using ground-based astronomical telescopes. As these telescopes continue to grow larger, this effect becomes more pronounced, demanding the use of real-time “adaptive optics”. The current era of adaptive optics comes with unique data challenges, including the processing of tens of thousands of noisy measurements to compute thousands of commands every, all within a couple of milliseconds. Traditionally, this real-time demand would only allow linear control laws to be employed, but the massive parallelisation allowed by CNN-based solutions is beginning to attract attention in the adaptive optics landscape as a non-linear alternative. CNNs (namely, the CGAN and UNet) promise a significant improvement to the quality of science achievable in astronomical instruments. Charles and Jesse will give a gentle introduction to the general problem statement in the context of adaptive optics, and an overview of their work in this domain.
This talk is now available on YouTube: https://youtu.be/85QFTGYAdss
Biographies
Jesse Cranney
Jesse Cranney began his post-doc at the Australian National University in 2021, where he is now leading the development of the adaptive optics control system for the next generation adaptive optics system: MAVIS, due to be seeing first-light at the ESO Very Large Telescope in Chile in 2027. Jesse is also responsible for the development of key wavefront sensing technologies for use in instruments of the Giant Magellan Telescope, currently under construction in Chile. Combining his experience in adaptive optics instrumentation with his passion for machine learning, it has been with great pleasure that Jesse has been able to collaborate on many projects at the cutting edge of both.
Charles Gretton
Since 2018 I have been convening the much celebrated and loved TechLauncher program at the Australian National University. That program comprises 35 teams of students working collaboratively and professionally together to: (i) create and leverage technology innovations to build and scale new enterprises, or (ii) fulfil statements of work and build relationships with business, industry, and government organisations in the Canberra region and around the world. I also lead projects on scaling logical inference with commonwealth government organisations, actively contribute to projects in the aerospace and astronomy domains, and occasionally collaborate with artist to expose and celebrate the clockwork orange in fashionable machine learning paradigms. In mid 2015 I co-founded Red Analytics PTY LTD (a.k.a. HIVERY), a Data61 spinout supported by the coca-cola founders platform. I was a senior researcher with the Data61 (NICTA) lab in Canberra from August of 2011. Prior to that I was a research fellow with the Intelligent Robotics Lab at the University of Birmingham 2008-2011; There I worked on a project investigating cognitive robots that could self-understand and self-extend. I was a researcher with the NICTA lab in Brisbane 2006-2008; There, I worked on fundamental research in AI Search, and simulation studies of city wide evacuation planning. I have also held adjunct positions at Griffith University since 2006. I am the primary author of NRMDPP, runner up in the probabilistic track of the 2004 International Planning Competition. I am a co-author of gNovelty+, winner of the "Random" category at the 2007 International SAT competition.
