Alberto Cruz

This book is a bit too simple. A lot of necessary topics in neural networks, such as differentiability of the loss function, are not in the book. Doesn't explain how the derivative of the loss function factors into the weight updates, students are left to discover this idea on their own, as to the reason why the step function perceptron and the sigmoid backprop weight update functions are different. The whole discussion on expert systems really needs to be factored into a FOL/Propositional logic domain. Needed things like existential quantifiers, universal quantifiers, and even how one would go about programming forward and backward chaining at the programmatic level are just glossed over. Highly recommend the Russel & Norvig book instead.