SMILe: Shuffled Multiple-Instance Learning