PhD position 12 – MSCA COFUND, AI4theSciences (PSL, France) - “Artificial intelligence at the...

"Artificial intelligence for the Sciences” (AI4theSciences) is an innovative, interdisciplinary and intersectoral PhD programme, led by Université Paris Sciences et Lettres (PSL) and co-funded by the European Commission. Supported by the European innovation and research programme Horizon 2020-Marie Sklodowska-Curie Actions, AI4theSciences is uniquely shaped to train a new generation of researchers at the highest academic level in their main discipline (Physics, Engineering, Biology, Human and Social Sciences) and master the latest technologies in Artificial Intelligence and Machine Learning which apply in their own field.

26 doctoral students will join the PSL university's doctoral schools in 2 academic cohorts to carry out work on subjects suggested and defined by PSL's scientific community. The 2021 call will offer up to 11 PhD positions on 12 PhD research projects. The candidates will be recruited through HR processes of high standard, based on transparency, equal opportunities and excellence.

Description of the PhD subject: “Artificial intelligence at the service of space astrometry: a new way to explore the solar system”

Context – Motivation

Astrometry is the discipline that aims to provide positions of celestial objects in space with the highest accuracy. Applied to planetary images, this provides information on the orbital and rotational dynamics of the objects studied, with sometimes fundamental consequences on the physical mechanisms acting within these systems (Lainey et al. 2009, Nature; Lainey et al. 2020 Nature Astronomy).

In this context, and while awaiting the arrival of the Juice mission (ESA) around Jupiter, the Cassini-Huygens (NASA-ESA) space mission will have been a great success. After thirteen years spent in orbit around Saturn, the probe will have collected several hundred thousand of images by the visible ISS (Imaging Science Subsystem) camera (Porco et al. 2004). Thus, a significant number of these images are now astrometrically reduced, allowing us to measure the position of Saturn's moons with an accuracy of typically a few kilometres (Tajeddine et al. 2013; Cooper et al. 2018; Zhang et al. 2021).

However, there is a significant probability that other moons, still undiscovered by classical methods, lie somewhere inside the large ISS dataset, their small size (up to a few kilometres) having prevented previous detection from Earth. Discovering new moons may shed light on the still debated formation history of the Saturnian system. Artificial intelligence (AI) approaches, and more particularly machine learning or recent deep learning techniques, would allow for a large-scale automated examination of the Cassini ISS data set, with the detection of new satellites and a precise reassessment of the position of the known ones as end goals. Another possible targeted objective is the automatic detection and classification of cosmic rays and their temporal evolution in the Saturnian system (related with the Saturnian aurorae, Solar activity...). In any case, the nature (size, geometry, contrast, etc) of the sought objects along with the need of properly shaped training data sets make their accurate detection a very challenging task. Several AI methods will be investigated, ranging from traditional machine learning with handcrafted image features to black box deep learning models comprising two-stages detectors (such as Faster R-CNN, Ren et al. 2016) and one-stage detectors (for instance YOLO and its variants, Redmon et al. 2016).

In any case, an IA-driven study of the ISS-Cassini images should help to better understand the current limitation in the analysis of the different images’ features, and how to overcome these issues. This will allow for potential important discovery while paving the way for the JANUS/Juice experiment (Palumbo et al. 2014).

Scientific objectives, methodology & expected results

Our scientific objectives rely on a systematic search for astrophysical objects present on the ISS/Cassini dataset, formerly not detected by classical algorithms. This shall provide exciting scientific results like new moon discovery and a monitoring of cosmic particles in the Saturnian vicinity. In addition, the thesis will provide a detailed methodology for investigating similar space images, like with the JANUS/Juice camera, exhibiting potential new capacity of the instrument.

Several tens of thousands of Cassini ISS images have already been reduced, providing a substantial set of labelled images including the position of Saturn's moons as well as stars in the sky background, whose position can also be inferred thanks to star catalogues such as GAIA (Brown et al. 2021). Moreover, we just finalized a new pipeline of astrometric reduction, which will be used over the whole set of remaining images, to provide us with more than 100,000 of images data. This constitutes an ideal basis for a deeper AI-driven examination of the whole data set. The recruited Ph.D student will therefore begin by an in-depth study of the literature related to object detection, both from a methodological standpoint (in particular the recent deep network architectures) and from an applicative standpoint to space images.

The object (satellites, stars, cosmic rays) detection and classification task will then be carried out, either in a unified framework or as separate tasks depending on the retained machine/deep learning solution. In particular, the cosmic ray detection task may require the generation of a synthetic cosmic ray data set, implying a good understanding of the physical interactions between a highly energetic particle and a CCD/CMOS sensor in order to generate realistic artificial images to populate the training data set. The Skybot tool (Berthier et al. 2016) and Gaia catalogue will also provide a baseline to which the obtained classification results will systematically be compared with. The exploitation and interpretation of the obtained results and their confrontation to the current understanding of the Saturnian system will constitute the next stage of the thesis. Last but not least, the student will attempt to generalize the contribution of the developed methods to other space missions, with the main target being the JANUS camera on board the Juice mission to be launched in 2022 (Grasset et al., 2013).

The Ph.D work will allow for several publications in refereed journals and international conferences, including a description of the AI algorithms used, a classification analysis of the different objects on ISS-Cassini images and their temporal evolution (e.g. cosmic rays), astrometric positions of outer moons of Saturn (previously undetected). Potential discoveries of new moons using AI will be emphasized with press releases.

International mobility

Our team has been working for years with C.D. Murray and N. Cooper (QMUL, UK) in the context of the Cassini mission and ISS data. The proposed work here is the evident next step toward ultimate precision from astrometric images. Hence, the selected PhD student will have the opportunity to go on interacting with the QMUL’s team in London.

The proposed thesis work is related to the Juice mission. Hence, the student will have the opportunity to visit the team of P. Palumbo (PI of the JANUS instrument) in Italy. Regular exchange in the context of the Juice mission and supported by CNES and IMCCE will be possible. She/He may participate to team meetings and present the results to the Juice sessions (among others) during international conferences.

Thesis supervision

Valéry Lainey and Guillaume Tochon

PSL

Created in 2012, Université PSL is aiming at developing interdisciplinary training programmes and science projects of excellence within its members. Its 140 laboratories and 2,900 researchers carry out high-level disciplinary research, both fundamental and applied, fostering a strong interdisciplinary approach. The scope of Université PSL covers all areas of knowledge and creation (Sciences, Humanities and Social Science, Engineering, the Arts). Its eleven component schools gather 17,000 students and have won more than 200 ERC. PSL has been ranked 38th in the 2021 Shanghai ranking (ARWU).