Generative Deep Learning: Teaching Machines to Paint, Write, Compose, and Play

How does a model learn to make something new? Generative Deep Learning answers that question by taking readers from neural-network fundamentals into the architectures behind machine-generated images, writing, music, and more. David Foster pairs technical explanations with practical examples in TensorFlow and Keras, making the book a working introduction to how generative models are built.

This second edition moves from core concepts to a broad survey of methods and applications. It is aimed at machine-learning engineers and data scientists, and offers technically curious readers a structured way to explore the ideas through code.

Start with the foundations, then build

The opening material introduces generative modeling, probability, representation learning, and deep-learning basics. From there, Foster develops the subject through model architectures, helping readers see how different approaches represent data and produce new samples.

A practical tour of generative models

• Variational autoencoders and GANs: explore approaches to generating and manipulating images.
• Autoregressive models and Transformers: examine sequence generation, including text and music.
• Normalizing flows and energy-based models: encounter alternative frameworks for generative modeling.
• Diffusion models: follow the process behind generating images through denoising.

Examples extend from training a GPT-style text generator to image-generation systems and polyphonic music. Rather than treating generative AI as a collection of finished tools, the book focuses on the underlying methods and how to implement them.

From core methods to wider applications

Later chapters turn to advanced GANs, world models, reinforcement learning, and multimodal systems. The closing section considers the development and wider implications of generative AI, including its applications and challenges. That breadth makes the book useful for readers who want both a technical foundation and a map of the field’s major approaches.

A book for learning by building

With its progression from fundamentals to modern architectures, this book is suited to readers ready to engage with machine learning concepts and code. If you want to understand what generative models do—and work through how they are made—Foster’s second edition offers a substantial, hands-on route into the field.