Alice’s Adventures in a differentiable wonderland: A primer on designing neural networks

by Simone Scardapane

Genres: Mathematics, Artificial Intelligence, Textbooks

378 pages, Paperback

Published July 16, 2024

Reviews

[Michiel · Rating 5 out of 5]

This was an excellent primer on connectionist models, ranging from simple regression to recent developments in transformers and LSTMs. Accessible yet fairly rigorous, I found it very useful to bring me up to speed with the essentials of modern AI.

[Zow Ormazabal · Rating 5 out of 5]

Even for readers already familiar with the topics covered, this book is an excellent resource for teaching. It presents all concepts with clarity, using clear and simple visual and code examples to support understanding. The text strikes a strong balance, explaining technical aspects with sufficient depth while remaining accessible to its intended audience: readers with some background in linear algebra and calculus who are beginning their journey into deep learning. It's quite approachable, and I will certainly recommend it to new students and researchers joining our lab.

The first part of the book can feel a bit slow, but the pace improves significantly around the 40% mark. The chapter on graphs could benefit from more practical examples and visual aids (which are heavily present in past chapters). Also, it spends a considerable amount of time on graph theory before introducing (graph) convolutions, which might discourage newcomers to the topic from continuing further.

It was a great read overall, and it felt entertaining and relaxing at the same time. I wish I had come across this resource when I started working with DL models!!

[Dan · Rating 4 out of 5]

Good content, but print version could have done with editing: poor phrasing, grammatical errors, choice of footnote fonts etc make it hard to read.

[Akshara Soman]

“a book that gets me hands-on as fast as possible while teaching the theory.

That's why I recommend this amazing book by Simone Scardapane.

More than 300 pages of high-quality explanations of linear layers, convolutions, transformers, recurrent networks, and auto-grad along with PyTorch and JAX code.” — umar jamil (LinkedIn)