Abstracts

1) Title: Artificial Intelligence techniques applied to the Solar Orbiter mission

S. Servidio, G. Lapenta, F. Valentini, J. Amaya, R. Dupuis, S. Perri, A. Retinò, F. Califano

European and international space missions have gathered a tremendous amount of observational data and there is now the need to extract from this growing data reservoir all important pieces of information. This goal can be achieved by the use of Machine Learning (ML) and Artificial Intelligence (AI) techniques. Very recently, a cross-disciplinary collaboration of scientists has successfully proposed a European project, named AIDA (Artificial Intelligence Data Analysis), which has been funded within the Horizon 2020 call (http://www.aida-space.eu/). The aim of the AIDA project is to make available to the general heliospheric community a new approach for the exploitation of data from past and planned European space missions, such as Solar Orbiter. The project makes use of Artificial Intelligence to extract information on events, processes and features from spacecraft data in a completely automated way. The amount of information that will be gathered from Solar Orbiter represents a unique opportunity for understanding fundamental phenomena that characterize the dynamics of the Sun and the interstellar medium.

2) Maching learning applied to SWA/PAS measurements - Separating protons and alpha particles

R. De Marco, R. Bruno, F. Marcucci, R. D’Amicis

Electrostatic analyzers are key instruments in space plasma measurements and, as such, they are part of the payload of several missions devoted to the study of the solar wind. They classify the ions according to their energy per charge ratio and incoming direction, so that their velocity distribution function can be defined. However, this kind of sensors can not distinguish between different type of particles. In solar wind case, where the major constituents are protons and alpha particles, the separation can be performed by analyzing the energy per charge spectra and applying specific fit procedures. We propose an alternative method based on the statistical technique of clustering. Clustering is a standard tool in many data-driven and machine learning applications. Our procedure proves to be useful in resolving tricky cases of overlapping particle distributions.