One option is starting to learn about DL from the big picture, then filling in the details as you go along, eventually going to higher math. I prefer this approach to the opposite where you first learn math foundation and then the implementation and use-cases. I liked the Fast.ai course.