Introduction to Deep Learning

Deep learning makes more sense when its building blocks are clear. In Introduction to Deep Learning, Jürgen Brauer starts with the biological neuron and the history behind neural networks, then develops the mathematical and computational ideas through models, code, and exercises.

The result is a teaching-oriented introduction that links neuroscience-inspired concepts to practical work with Python and TensorFlow. Its progression gives readers room to understand how the models are constructed before moving on to convolutional networks and training techniques.

From biological neurons to learning models

The early chapters examine the structure and signaling of biological neurons, then consider how artificial neuron models can function as feature detectors, filters, and components in larger networks. This groundwork leads into perceptrons, self-organizing maps, and multilayer perceptrons, including discussion of learning, gradient descent, backpropagation, and model limitations.

Neural networks, built and examined

Python examples and TensorFlow chapters bring the concepts into a programming context. The book introduces convolutional neural networks, including convolution and pooling layers, their parameters, and output dimensions. It also addresses challenges and methods used in training, from vanishing gradients to momentum optimization and batch normalization.

More than architecture

The closing chapters look toward ideas beyond conventional deep learning, including attention, lifelong and incremental learning, embodiment, prediction, and cognitive architectures. The book’s exercises reinforce the material with work on Python, perceptrons, self-organizing maps, backpropagation, TensorFlow, and CNN experiments.

A structured starting point for deep learning

This first edition is suited to readers who want to study neural-network foundations alongside implementation examples—not just encounter a list of model names. Students and independent learners with an interest in machine learning can follow the progression from neuron models to deeper architectures and use the exercises to revisit the key ideas in practice.