I really enjoyed the section on symbolic data in SICP https://xuanji.appspot.com/isicp/2-3-symbolic.html - gives lots of good insight to start thinking about how automatic differentiation is implemented in the deep learning libraries.
FWIW, automatic differentiation (as used in popular machine learning frameworks) is not the same thing as the symbolic differentiation described in that section.
HtDP is published by the MIT Press, but I'm not sure that's the same as saying it's "from MIT". None of the authors of HtDP are at MIT. One is at Northeastern, one at Northwestern, one at Utah, and one at Brown. Lots of CS books are published by MIT Press regardless of whether MIT actually makes use of the material.
Also, HtDP uses PLAI, a language implemented in Racket, which is not pure Scheme, but that's a pretty minor nitpick that doesn't really detract from your point. HtDP is a pretty good textbook, in my (admittedly limited) experience with it.