1

I am confused with the definitions of domain adaptation and multi-task learning.

I have K datasets, each with the same feature and label space and thus the same learning problem, but with different domain P(x,y). For each dataset, I would like to learn a model that utilizes information from the other datasets. Would the appropriate technique be to use multiple-source domain adaptation K times on K different target domains or to use multi-task learning one time?

I have read different definitions of multi-task learning and am confused whether MTL is only useful for tasks with different feature and label spaces.

Tim D
  • 11
  • 3

1 Answers1

0

I am in no way an expert or even a newbie to the field. But from what I have read from the web and other resources, I can say that there isn't a proper set definition for multi-task learning and even meta-learning to a certain extend. So, to answer your question with very very limited knowledge, I would say that your problem statement can fall into either of those two fields. However, there is one way to differentiate between the two. Both the problems can be considered as solving distinct tasks which share some similar underlying structure. Now the tasks which have similar underlying structure but look like two distinct tasks to us, wrt to data distribution and modality, then those tasks might fall into multi-task learning. And the task which don't look quite distinct, wrt to data distribution and modality, they might fall into domain adaption. I know the answer is quite vague and subjective, but the answer is bottlenecked by my limited knowledge in the field.

Hetansh Patel
  • 1