Combining multi-task learning with multiclass classification

Is it possible to combine both of them or this type of learning is mutually exclusive to binary classification?

Let’s say I want my model to learn from a video and get 3 outputs but one of these outputs has 4 classes for classification.

Hi, @Lujain_Andijani !

Sure! You can always combine the backbone you want to a multi-head output and tweak each of those heads to be however you want (one-hot encoding or another type for classification)

1 Like

Thank you so much! I find this advice to be very helpful, instead of making 8 models for the whole project that’d be shortened to 3 for the project I’m working on.

I hope this helps. Beware that multi-output training is not trivial and you may have to deal with many different settings that are not that important in normal one-headed networks. Check this paper if you want

1 Like