It depends on what you want to do. Training and testing models with deep learning doesn't require PhD. Designing network architectures, debugging and pushing mo

It depends on what you want to do. Training and testing models with deep learning doesn't require PhD. Designing network architectures, debugging and pushing model to its limits requires significant insights in to theoretical foundations on the other hand. Doing something novel like GANs would require even more formal background. However, I would say anyone with undergraduate math (linear algebra and bit of calculus) can learn all of this if they are determined and willing to invest couple of years of intensive study time. PhD programs just facilates this more easily along with certificate of approval from experts in the field that others can trust about your skill sets.