Neural Networks and Deep Learning — Michael Nielsen

[needs image URL — source: http://neuralnetworksanddeeplearning.com/]

A free online book that teaches the core ideas of neural networks and deep learning from first principles — and a model of open, freely readable science writing.

Type: book (free online)
By: Michael Nielsen
When: 2015 (developed in beta from 2013; freely hosted online)
Where it sits in their arc: the landmark
Where to get it / join: http://neuralnetworksanddeeplearning.com/

What it is

Neural Networks and Deep Learning is a free, openly licensed online book that builds an intuition for how neural networks work, why they are hard to train, and how deep learning makes them tractable. Rather than racing through formulas, Nielsen develops each idea slowly with worked examples and interactive diagrams, so a motivated reader with modest math can genuinely understand backpropagation and the principles behind modern networks. It became one of the most widely recommended on-ramps to the subject precisely because it teaches understanding rather than recipes — and because it is given away.

Core ideas

  • Understanding over recipes — The book optimises for the reader actually grasping why a technique works, not just how to call it.
  • First-principles pedagogy — Backpropagation, gradient descent, and overfitting built up step by step with intuition before notation.
  • Open by default — Free to read, openly licensed, supported by readers — Nielsen’s open-science values applied to teaching.

How it connects to the Guild’s practice

This book is a case study in two of the Guild’s commitments at once: teaching for genuine understanding, and sharing knowledge openly rather than gating it. It sits in the learning / memory lens (see The Disciplines — Many Lenses, One Room) and pairs naturally with Nielsen’s later mnemonic-medium work — here he gets you to understand a hard subject; in Quantum Country he tackles the next problem of making sure you remember it.

Related works

Notes from the room

Space for members to add takeaways and how they used it.