Found this to be an interesting survey of design patterns for CNNs: 14 Design Patterns to Improve Your Convolutional Neural Networks.
Very high level, but with references, and it reads like a list of things to try or investigate further.
I was interested to see several points from this course are mentioned, e.g. fine-tuning an existing model, normalizing inputs, etc. I’m still just a n00b at DL, so all this stuff is new and shiny to me.
I did a search on this forum and the wiki and didn’t find this link, so maybe this will be new and helpful for others too.