How Deep Learning Works. An Experiment-First Guide for Programmers

A practical, experiment-driven introduction to deep learning that shows programmers how modern neural networks actually work by running, inspecting, and reasoning about real models, from foundational concepts through large language models. How Deep Learning Works is for programmers who want to understand deep learning, not just use it. Rather than starting with heavy math or abstract theory, this book takes an experiment-first approach.
Each chapter walks readers through carefully designed experiments using real deep learning toolkits, guiding them step by step from running code to understanding why models behave the way they do. A consistent structure-overview, execution, code review, and discussion-keeps readers oriented and focused on building intuition, not memorizing APIs. Beginning with fundamental ideas like classification and neural networks, the book steadily builds toward modern techniques, including transfer learning, zero-shot and few-shot models, and hands-on experiments with large language models.
Along the way, readers learn how to spot failure modes, reason about trade-offs, and adapt existing tools to new problems. The result is a clear, software-centric explanation of deep learning that helps working programmers move beyond copy-paste ML and develop real understanding they can apply in their own projects.