How Neural Networks Work
A Gentle Journey from Perceptrons to CNNs, RNNs, and Beyond
Build a real intuition for neural networks from the ground up. Start with a single perceptron, work through activations, losses, backprop, and optimisers, then build CNNs for vision and RNNs for text — all with Python first and PyTorch second so the ideas stick.
Foundations— What neural networks are and the tools you'll use.
What Are Neural Networks?
The big picture of neural networks
Python and PyTorch Essentials
Setting up your toolkit for neural network development
Mathematics for Neural Networks
The essential math you need and nothing more
Building Blocks— Perceptrons, activations, and losses.
The Perceptron
The simplest neural network
Activation Functions
The non-linearity that makes networks powerful
Loss Functions
How networks measure their mistakes
Training a Network— Forward pass, backprop, and optimisers.
Forward Propagation
How data flows through a network
Backpropagation
The algorithm that makes learning possible
Optimizers
Smarter ways to update weights
Deep Networks— MLPs, training in practice, and regularisation.
Multi-Layer Perceptrons
Stacking layers for more power
Training in Practice
Everything you need for real training
Regularization
Preventing overfitting
CNNs for Vision— Convolutions, architectures, and an image classifier.
Understanding Convolutions
How networks see images
CNN Architectures
Classic and modern CNN designs
Image Classification Project
Build a complete image classifier
RNNs for Sequences— RNN, LSTM/GRU, and a text classifier.
Recurrent Neural Networks
Networks that remember
LSTM and GRU
Solving the memory problem
Text Classification Project
Build a sentiment analyzer
Going Further— Modern training techniques and debugging.
Modern Training Techniques
Tips and tricks from the pros
Debugging and Improving Networks
When things go wrong and how to fix them
Where the book lands in practice.
65 sections. Begin with one.
Chapter 1 — What Are Neural Networks? — is where every reader starts.