Where and when

speaker: Sabrina Guastavino
affiliation: Dipartimento di Matematica, Università di Genova
date: September 30^th, 2022
venue: Athens (Greece) or on-line (Microsoft Teams), SWATnet Workshop 3 organized by Supervisory Board of the SWATNET EU project

Machine- and deep learning in solar flare forecasting

abstract: Solar flares are the most explosive phenomena in the heliosphere, releasing a huge amount of electromagnetic radiation at all wavelengths and, in this way, triggering the whole space weather connection. Solar flares originate from magnetically active regions (ARs) on the Sun. However, not all active regions give rise to solar flares and the nature of the prediction is intrinsically probabilistic. In the last decade machine and deep learning approaches have been obtaining an increasing interest in flare forecasting, thanks to flexible algorithms that may take as input point-in-time feature sets extracted from magnetograms, time series of features, point-in-time images of ARs, and videos of magnetograms of ARs. The prediction performances of these supervised approaches are characterized by a notable degree of heterogeneity, which is probably related to significant differences in the way datasets are generated for training and validation.
In this talk we discuss some crucial aspects for the forecasting effectiveness of machine/deep learning for flare prediction: (1) the optimization of the network weights which depends on an appropriate choice of the loss function in the training phase; (2) the validation strategy for the prediction method and (3) the assessment of the prediction power of the method. Finally, we show a validation strategy based on the generation of well-balanced training and validation sets and we discuss its performances by means of a video-based deep learning approach.

SWATNET Workshop 3 / Solar Activity and Space Weather: Physics Behind the Process
AI-based Modeling of Solar Eruptions: Machine- and Deep-Learning

https://swatnet.eu/workshop-3-solar-activity-and-space-weather-physics-…