Kind of an aside - what do you need to learn to create "deep dream" style generative programs, and can you learn it from fast.ai?
The keywords you're looking for are generative adversarial networks. Not sure if the syllabus has it.
That technique is used for a lot of AI generated images, but Deep Dream specifically used a trained recognition network "backwards" to adjust the image.
It's like a generative network without the adversarial part.