EPIC & NNExI

In 2019 the PSILab has tarted two new projects ; the ERC consolidator grant EPIC and the ULiège-funded ARC grant NNExI.

• ERC Consolidator grant EPIC (Earth-like Planet Imaging with Cognitive computing) aims to leverage the power of machine learning techniques to optimize the scientific exploitation of vortex coronagraphs, in particular in the context of METIS, one of the first-light instruments for the Extremely Large Telescope (ELT).
• ULiège-funded ARC grant NNExI (Neural Networks for Exoplanet Imaging) aims to make fundamental advances in the field of machine learning, including likelihood-free inference techniques, with applications in image processing and focal-plane wavefront sensing. This project is carried out in collaboration with Profs Gilles Louppe and Marc van Droogenbroeck from the Montefiore Institute (ULiège).

Within these two projects, we have started a series of research programs, along the following lines:

1. Application of deep learning techniques to exoplanet imaging (in collaboration with Profs Marc Van Droogenbroeck and Gilles Louppe, Montefiore Institute, ULiège). Description: Identifying the image of faint exoplanets around bright stars while observing through the turbulent Earth atmosphere is a challenging task, which requires subtracting a model of the stellar point spread function by image processing. Over the last ten years, building this PSF model has evolved from a pixel-wise median to a principal component analysis, and now to machine learning techniques (Gomez Gonzalez, Absil, & van Droogenbroeck 2018). Here, we propose to develop and test new algorithms based on machine learning techniques, focusing particularly on the joint estimation of planetary flux and position as an inference task. We will also test the possibility of training discriminative models based on synthetic data, which will require improving a generative model already available within the team. In parallel with these ground-based applications, we will investigate how these techniques can be adapted to the case of space-based observatories, focusing specifically on the MIRI instrument for the James Webb Space Telescope. Finally, we are planning to reformulate the analysis of High Dispersion Coronagraphy data sets (combining high-contrast imaging with high-resolution spectroscopy) within a supervised machine learning framework.
2. High-contrast imaging data challenge (in collaboration with Dr Gomez Gonzalez, BSC, and Dr Faustine Cantalloube, MPIA). Description: Recently, we have launched an exoplanet imaging data challenge to compare the merits of various image processing algorithms. In the coming months, we will tackle the assessment of the data challenge entries, and focus on the development of robust performance assessment tools. In the coming years, we will use that experience to develop a more ambitious version of the data challenge.
3. Scientific exploitation of high-contrast imaging instruments. Description: Through our collaborations with various observatories around the world, we have access to high-quality high-contrast imaging data sets. Within the EPIC/NNExI projects, we will continue to contribute to their scientific exploitation. In particular, we are participating to the analysis of the 100-h worth of mid-infrared HCI data obtained within the NEAR project with the VISIR camera at the Very Large Telescope. We participate to several observing programs at Keck, LBT and VLT, using the vortex coronagraphs that we have developed within the VORTEX project. We will also contribute to the early scientific exploitation of the high-contrast imaging modes of the ERIS camera, to be installed at the VLT in 2020.
4. Development of focal-plane wavefront sensing techniques using machine learning. (in collaboration with Prof Gilles Louppe, Montefiore Institute, ULiège). Description: One of the main limitations to the performance of current and future high-contrast imaging instruments comes from the non-common path aberrations (NCPA) appearing between the wavefront sensor used to correct atmospheric turbulence in the adaptive optics system and the coronagraphic instrument itself. In order to reach the ultimate sensitivity of these instruments, NCPA need to be measured and corrected. One of the most promising methods to do so is to infer the wavefront aberrations based on the scientific images obtained by the coronagraphic instrument. Here, we propose to apply the most recent advances in machine learning, including invertible neural networks and likelihood free inference, to reconstruct the wavefront based on focal-plane images, augmented by contextual information from various sensors. Training the networks will be based on a generative model already available within the team. The newly developed algorithms will then be tested in the lab and/or on sky, in collaboration with a postdoctoral researcher working within Dr Absil's group.
5. Development of next-generation vortex coronagraphs (in collaboration with Prof Mikael Karlsson, Uppsala University). Description: During the last few years, we have developed, manufactured, and tested vortex phase masks based on concentric subwavelength gratings etched onto diamond substrates. These vortex phase masks have been installed in some of the world-leading observatories (VLT, Keck, LBT). With this project, we aim to improve the performance of vortex coronagraphs by (i) developing, manufacturing, and testing new vortex phase masks with higher topological charge, which are more resilient to low-order aberrations, (ii) designing and testing custom apodizers tailored to specific telescope pupils. Part of this work is specifically targeting future applications on the European ELT, within the METIS project.

Within the EPIC/NNExI projects, we also actively contribute to the development of the METIS instrument for the ELT. This part of the project is described in more details on the METIS page.