Lead: UNIGE
Objectives
Research most appropriate deep learning (DL) architectures for each ARCAFF objective.
Update and or development of neural network (NN) architectures for each ARCAFF objective if necessary.
Train, optimise, and evaluate DL models for each ARCAFF objective.
Analyse trained DL models looking for physical insights.
Deliverables
D3.1: Trained active region localisation and classification deep neural networks and software (Month 18)
Release of the training pipeline and model software as well as the trained model weights for active region (AR) classification and AR localisation and classification objectives.
D3.2: Trained point-in-time magnetogram and multimodal flare forecast deep neural networks and software (Month 26)
Release of the training pipeline and model software as well as the trained model weights for the point-in-time flare forecast objectives, both magnetgram and multimodal.
D3.3: Trained time series multimodal flare forecast deep neural networks and software (Month 36)
Release of the training pipeline and model software as well as the trained model weights for the time series flare forecast objective.