Neural networks are all around us, from language models to robotics, and understanding them involves delving into differentiable programming. This primer is designed for beginners like Alice, offering an intuitive introduction to optimization, automatic differentiation, and common design techniques. The goal is to bridge the gap between theory and code, focusing on convolutional, attentional, and recurrent blocks. The book covers a range of topics, including linear models, transformers, graph layers, and more advanced material, with additional chapters available for feedback. The author hopes to eventually release a second volume but acknowledges that it may never happen.
https://www.sscardapane.it/alice-book