Python Neural Network Programming

This book is an introductory guide to neural networks by the British author Tariq Rashid, translated into Chinese by Lin Ci and published by Posts & Telecom Press as part of the "Deep Learning Series." Rather than piling on formulas from the start, it begins with the simplest ideas and explains step by step why neural networks work. Chapter 1, "How Neural Networks Work," starts from a simple predictor and, using concepts such as classifiers, multi-node weights, and error backpropagation, breaks the mathematics of gradient descent and weight updates into small, traceable steps, then introduces matrix multiplication so readers can understand how a three-layer neural network passes and returns signals internally. Chapter 2, "DIY with Python," turns to hands-on practice: it first introduces the basics of Python and IPython, then builds a neural network by hand in Python and trains and tests it on the classic handwritten-digit dataset MNIST to check its recognition performance. Chapter 3, "Fun and Games," takes the trained network out of the black box: experimenting with your own handwritten digits, inspecting the network's internal state, expanding the training data by rotating images, and, in the closing remarks, discussing ways to further improve performance. Appendix A covers the basics of calculus, and Appendix B demonstrates how to install IPython on a Raspberry Pi and train and test a neural network. Concise and richly illustrated, the book focuses on building intuition and hands-on ability rather than cataloguing the usage of deep learning frameworks, making it a good first book for understanding the principles of neural networks.