https://deepmind.com/blog/article/traffic-prediction-with-advanced-graph-neural-networks
Traffic prediction with advanced Graph Neural Networks
“We discovered that Graph Neural Networks are particularly sensitive to changes in the training curriculum… the most successful approach to this problem was using MetaGradients to dynamically adapt the learning rate during training… making use of a linear combination of multiple loss functions (weighted appropriately) greatly increased the ability of the model to generalise… we are exploring whether the MetaGradient technique can also be used to vary the composition of the multi-component loss-function during training…”