What is a good way to learn to play baseball? Is it by starting to learn about the physics of a parabola and learning how to stitch a ball, or by watching baseball and tossing the ball back and forth? How does this apply to learning deep learning (or anything else)?
Just as we would not teach how to play baseball to a kid (or anyone for that matter!) by teaching them calculus first, we will not learn deep learning in this fashion either! Instead, in the course, we will learn by playing the whole game - learning how to build end to end models that work! Only gradually, after we understand the big picture, will we start going deeper.
Relevant part of lecture
Making Learning Whole - a book by David Perkins whose research results are the basis for the method of teaching adopted in the fast.ai courses.