Research: artificial intelligence required in the field of astronomy!
In March, the teams from the EPITA Research and Development Laboratory (LRDE), Institut de Mécanique Céleste et de Calcul des éphémérides – Institute of celestial mechanics and ephemeris calculations (IMCCE) Paris Observatory and the Physics and AstroPhysics (PAP) research team from IPSA (already working in collaboration with IMCCE) met at the school’s Parisian campus for a joint seminar. This event marked the beginning of a brand new collaboration between the researchers from the two entities, focused on artificial intelligence and image processing to benefit astrometry. To discuss this unique link between computer science and space, EPITA spoke with Valéry Lainey, an astronomer at the Paris Observatory, and Guillaume Tochon, a teacher-researcher at the LRDE and co-director of the IMAGE major.
Guillaume Tochon and Valéry Lainey
Valéry Lainey: I have been an astronomer at the Paris Observatory for the past 20 years and mainly work on the dynamics of planets and their natural satellites, such as those of Jupiter and Mars. I’m extremely interested in the physics of these systems, allowing me to better understand them. My research focuses on the formation of these systems, their long-term evolution, their internal structure, their physical properties… And I am also very interested in space missions. Indeed, Europe will send a space mission named JUICE to visit the Jupiter system in 2022. It is a very ambitious mission that will surely fascinate the population, while helping us acquire new knowledge about the system. As I am highly involved in this mission, I am extremely focused on all data, observations and imaging. My work includes a dynamic aspect on Newton’s equations, to put it in simple terms – and an observational part – we have images, and we strive to measure the position of the objects we are interested in, with the greatest possible precision.
Guillaume Tochon: As for me, I have been a teacher-researcher at the LRDE for a little over four years. Before that, I wrote a thesis on satellite imagery in Grenoble, and I have been working in the field of image processing for about ten years. Until now, I have mostly worked on images of the Earth – with the sensors pointing “downwards” -, but I always wanted to work on images “from above” in space… And finally, last summer, with Élodie Puybareau, a colleague from the LRDE and one of Valéry’s colleagues, we oversaw the internship of an EPITA student, who had to detect and classify images of meteors taken from the ground. That’s how I started working with applications related to astronomy and began increasing the number of contacts with the Paris Observatory teams.
Guillaume: We all officially met to present the subject matter we were respectively working on and discuss possible collaborations with the Paris Observatory, including researchers from the PAP team at IPSA, and the LRDE. There are a lot of data to process at the Observatory, but they lack in image processing and artificial intelligence systems; whereas, for the LRDE, it is important to work on concrete data, which is very exciting. It was therefore important to meet up and exchange ideas in order to decide upon future collaborations.
Valéry: For the record, as I am also the head of a team called Pégase at the Paris Observatory laboratory, my initial conversations with Guillaume concerned a much more modest artificial intelligence goal. Like for all the new “popular” technological trends, as researchers we wanted to learn about them so that we wouldn’t be lagging behind in a few years and also because we feel it is important for our work. We thought that we could simply attend a seminar on artificial intelligence led by the LRDE, so that our team would be up to date on this issue. However, while discussing the details of this seminar, Guillaume and I quickly realized that there were many other possibilities for both of our teams. We were all excited and that’s how the idea for this first seminar came about. Not surprisingly, this meeting demonstrated that we could collaborate on several exciting research projects together!
Guillaume: That’s right. When we create artificial intelligence, what we need first of all are data and it doesn’t really matter what application we use in the end. However, even if those working at the LRDE or the Paris Observatory do not belong to the same communities, we speak the same language as researchers and share the same desire to publish our findings and make a contribution to our field. And now, we have the opportunity to bring two communities together to work on special projects, which is extremely interesting! We bring our knowledge, and in return, we receive new knowledge. It’s a win-win situation.
The teams from the two laboratories at the seminar
Valéry: There are two types of data. The first involves images of space – black and white images with white glowing dots representing celestial objects. What interests us is measuring the “pixel positions” of the objects in the image. It sounds simple, but these objects are difficult to see. Indeed, if you have a white object on a black background, it is easy to spot, but this is not the case when you have a gray object on a slightly less gray background – and this is very important as it can give us a great deal of information! The other data concern another activity: photometry. This consists of observing how the intensity of a light source varies over time. In this case, we are faced with a one-dimensional curve, showing the decrease in flux over time.
Guillaume: The contributions made by artificial intelligence are based on two aspects. The first is image processing. At LRDE, we are experts in the field of mathematical morphology image processing: this is a set of very robust image filtering and analysis techniques that lend themselves to gray scale images, such as those used by Valéry’s team. The second aspect concerns the large volume of data to be processed: this is what makes artificial intelligence so interesting. We must teach a model good behavior, which can then be replicated to easily automate the processing of large volumes of data.
Guillaume: Absolutely. There is an “output” behavior determined by the end use and we have to translate that behavior into rules given to the model. Because of this, we need to fully understand what the output is to calibrate it.
Valéry: From this point of view, we bring our expertise as we are perfectly familiar with this data. I think Guillaume and the LRDE researchers appreciate working with people are fully familiar with their data, know what they are looking for and can target flaws in the algorithm. From the very beginning, we have the information we need and know what direction we want to go in as well as what our weaknesses are: this allows for the optimal collaboration on an artificial intelligence project!
Guillaume: We can interact on several levels. On the one hand, with collaborations solely between researchers, and on the other hand, collaborations related to supervising students, trainees, or even long-term PhD students.
Guillaume: This is still one of the possibilities, although nothing has yet been established. However, we have already discussed the possible implementation this year of a PFEE (an end-of-studies project generally in connection with a company), focused on applications useful to Valery’s team and the Paris Observatory!
Get the latest news about LRDE on its website and LinkedIn.
Learn more about IMCCE by visiting its website.