Efficiently clustering tasks based on higher-order task affinities improves multitask learning performance.
