Uncertainty-based loss function weighting for multi-task learning (Kendall et al., 2017).