search

FrankWork / fudan_mtl_reviews

TensorFlow implementation of the paper `Adversarial Multi-task Learning for Text Classification`
178 stars 40 forks source link

Loss linear combination #12

Open gabrer opened 5 years ago

gabrer commented 5 years ago

Are the coefficients used while summing up the losses fixed? How did you choose them? In the original paper, they would be _alphat, where t is the particular task.

Thank you!

FrankWork commented 5 years ago

I think it is a hyper-parameter. I choose one with higher F1 score.

gabrer commented 5 years ago

Ok, thank you! Still looking for any other approach.