What courses do you want to see the deeplearning.ai team build next?
A course on hyperparameter and architecture search would be a topic interesting enough to take. In particular, implementing GA for both hyperparameter and architecture search and using multiple GPUs for distributing search. Advanced HPC concepts for using GPU clusters to distribute training, using a parameter server, and performing distributed search are also interesting (but understandably expensive). However, a toy problem that runs quickly would still give hands on experience and build confidence for learners who want to utilize these resources.
Seconded! I'd also like to see more practical hyperparameter searching methods explained, distributed trainting, etc. Building on that ask, I think I'd make my request more general and ask for a course on "Deep Learning in the Wild", like practical skills for when you actually use deep learning. I think we come to these courses because we want to hear from people with experience (Andrew and Company) because we respect their opinion and so we don't need to reinvent the wheel. I, and I assume "we", are all looking for "The stuff we can't learn from a textbook or blog post."
Thinks like setting up your local and/or cloud environment, GPU setup, practical architecture design, hyperparameter search beyond random, Distributed DL paradigms and set up, reading from "Big Data" distributed file systems, how to get unstuck when you encounter a problem that isn't easily fixed by reading Stackoverflow/documentation, maybe even project design (how to ask the right questions before you even get started?).
I'd propose a capstone or final project to put these skills to use, and the course video content could be some of these commonly, often unaddressed and yet common questions. It's easy enough for us to complete the spoon fed assignments in the coursera specialization, and while it's useful to build some of these things from scratch, we're not actually getting a good feel for what the process and strategy experts use to approach a new problem.
* +1 for Reinforcement learning
* (Semi-) unsupervised NNs
* Financial markets and NNs! While it sounds specific, in sci. papers it leverages so many advanced NN techniques from the every course in the spec and beyond, so it may be a great vehicle for a course on advanced subjects and integration of the entire specialization. Plus a hot subject by itself 🙂
Reinforcement learning, unsupervised learning in all its flavours and applications by TensorFlow shoulb be very appreciated.
I think the style used in DL courses of Andrew Ng is a perfect mix of theory and practice and should be taken as an example.
In my opinion in every course more time should be spent on Keras and TensorFlow exercises because they've not a friendly API interface and are very tough to master!