Using CNN for different purposes

I have read in many blogs and talks that CNN is going to be the new trend, i.e., deep learning in non-images. I was wondering how can we apply it for say problem set like titanic? (Specifically how would we handle categorical variables (one-hot encoding (?)) )