Using SimMTM for feature extraction

[Bili-Sakura]

Awesome work!

I am working with a time-series data for regression (feature extraction/representation learning) task, where

 X \in R^{S \times L},  Y \in R^{S}

and we model $f_\theta(X)=Y$.

I wonder is it acceptable that we direct use the backbone of SimMTM for Classification and replace the classifier-head with our target output mapping module. I will appreciate if you give any suggestions.