any strategy of assigning the loss weight to different loss function in multi-tasking learning?

[helxsz]

I am training a multitasking network on face detection and face pose detection (5 different pose), normally the weight for the loss function can be equal for the two tasks, but I just wonder any optimization or systematic way giving the strategy of assigning the weight to different task?