Differentiable Modelling, dLux, and the next generation of scientific modelling tools

Louis Desdoigts

The University of Sydney

Software tools and computational methods are starting to become recognised as fundamental pillars in the future of astronomy. Despite this, the majority of softwares used throughout the community do not take advantage of the key piece of technology that has enabled the revolution in machine learning: automatic differentiation or "autodiff". This talk presents dLux, an open source optical modelling software utilising autodiff, and explores how this unlocks a host of new and powerful methodological approaches for instrumental design, calibration and analysis. It further examines how these methods have been applied to calibrate detector non-linearities in the JWST Interferometer, revealing previously unknown data artefacts and providing the most precise calibration of the instrument to date.

Date:	Thursday, 16 May 2024
Time:	11:30
Where:	All
	Pavillon MIL local A-3521