François Chollet, the creator of Keras, and Tomasz Kalinowski, a Posit software engineer who maintains the Keras and TensorFlow R packages, teach deep learning from first principles using R. The third edition is built on Keras 3 and treats R as a fully capable environment for modern deep learning, so you can develop models end to end without switching to Python.
The book is aimed at readers with intermediate R skills and no prior deep learning experience. Each chapter builds intuition step by step through carefully constructed code examples and clear illustrations, with an emphasis on understanding why models work and not only how to use them. This expanded edition goes well beyond the fundamentals, with new chapters on the transformer architecture, building your own GPT-style large language model, and generating images with diffusion models. Experienced practitioners will also find useful perspective on the nature of deep learning.
Along the way, you’ll practice:
- Understanding deep learning from first principles
- Working with the latest features of Keras
- Classifying and segmenting images
- Forecasting time series
- Classifying text and translating between languages
- Fine-tuning pretrained models and working with transformers and large language models
- Generating text and images, including with diffusion models
- Tuning and scaling models
The source code for the book’s examples is available on GitHub. The second edition, based on the previous version of Keras, covers partly different material.
