- Is Machine Learning required or is useful for understanding (theoretically and practically) Deep Learning?
Deep learning is itself a huge subject area with serious applications in NLP, Computer Vision, Speech and Robotics.
You should learn deep learning from scratch like understanding forward propagation, back propagation, how weights are updated etc.. instead of using high level frameworks like keras, pytorch.
It's OK to use them once you understand the basics to save time and code complexity, but remember "surely" you don't need machine learning for that.
Since you are familiar with the mathematics part, I would suggest you to straight away jump into Deep Learning. Note that deep learning is inspired by how the brain works.
- Can I start right away with Deep learning or should I cover Machine learning first?
Yes you may start right away, start with the hello world problems "MNIST DIGIT Classification" if you know little image processing.
Start with a simple neural network model from scratch, then use keras (very easy) and then proceed to CNN ...
You may start with simple problems in other fields too (NLP, Speech)
I suggest Andrew Yangs, course in Machine Learning (within this he explains a neural network model for MNIST I guess).
- In what way machine learning useful for Deep learning?
You will understand that in machine learning you sit down and find useful features in the dataset yourself, but in deep learning it happens automatically
(Learn Deep learning in detail and come back and read this, you will understand exactly what I mean!)
If you learn Machine learning and then go to deep learning, you will realise that it was unnecessary .if you interested in this field of AI, jump into deep learning right now!