How to examining the training curves about a multi-task learning model?

[JohnHerry]

I had read the part about how to read training curves, Is that questions still ok for a multi-task training pipeline?

[Varun Godbole]

Yeah our advice should still be relevant for a multi-task setting. Often in multi-task settings, you can make your life easier if you come up with a single loss that is a function of all the different tasks. And attempt to directly optimize that.

[JohnHerry]

Thanks for the help！