Differentiable Optics and Exoplanet Direct Imaging
I am very interested in applying differentiable optical models to exoplanet direct imaging and beyond. In our ∂Lux project, we use automatic differentiation in JAX to simulate wave optics propagation, and invert this with Bayesian inference for phase retrieval and instrument design.
We have built amigo, a digital twin of the JWST Interferometer incorporating a ∂Lux model of the optics and a neural network effective detector model. We have achieved photon-noise-limited calibration and very high fidelity image reconstruction with AMI, and we are now working on coronagraphy and widefield imaging problems.
My Honours and Masters research was on kernel-phase interferometry, working with Peter Tuthill (Sydney) and Frantz Martinache (Observatoire de la Côte d'Azur). This generalizes non-redundant aperture masking to construct self-calibrating, closure-phase-like quantities from redundant apertures. We applied this to detect faint companions from ground-based AO and space telescope data, and also for wavefront sensing.
I have long been interested in broader extensions, proposing kernel amplitudes, showing that coronagraphs possess a kernel, and "Delay-Insensitive Self-Calibrating Observables" (DISCOs) that extend to the Fresnel regime.
