In case its of some use to anyone… Click here for pdf.

This document demonstrates the mechanics, underlying a simple but not completely simplistic fully connected neural network (NN). In particular the multivariable mathematics involved in backpropogation is illustrated with a running example, which eschews the irritating handwavy approach found in many books, all of which seem to assume you can’t differentiate (some of these texts do cover differentiation but they assume the single variable case, while even simple NNs are multivariate in real life). Click here to view (Gave up trying to persuade Jekyll Markdown to display Mathematics).

Blog post related to ConcreteManifolds Repo. Click here to view (its difficult to persuade Jekyll to include a complex HTML page with JavaScript dependencies).

Personally I find Android Studio Clion based C++ editing somewhat irritating as it does not seem to integrate with even slightly complex CMakeLists.txt configurations properly leading to correctly specified include files being flagged as missing. This normally results in particularly annoying popup error windows which don’t go away and get in the way when you are editing.