When is an SHM problem a Multi-Task-Learning problem?