Non-expert DL practitioner here. The coverage is very good - if I were running some kind of "Deep Learning onboarding class" for a university or tech company, t

Non-expert DL practitioner here. The coverage is very good - if I were running some kind of "Deep Learning onboarding class" for a university or tech company, this is what I would present. My favorite aspect of this book is that it provides a graphical models interpretation of DL methods, which is the most powerful perspective we have right now to reason about model design (instead of some large black box function that we train end-to-end without knowing what's in between). It also explains some fairly recent models and techniques well (VAE, DCGAN, regularization) that form the basis of more complex architectures. If you understand the models here, you should be able to understand the design choices made in more complex architectures. Thanks to Goodfellow, Bengio, and Courville for this excellent work.