§1.1–1.2Who Should Read This Book? · Historical Trends

Before the history, the book closes its opening argument with two famous figures that place deep learning deep learning Machine learning via representations expressed in terms of simpler representations — a deep concept hierarchy. defined in ch. 1 — open in glossary on the map of AI disciplines.

Where deep learning sits (fig 1.4)

Deep learning is a kind of representation learning representation learning Using learning to discover not just representation→output but the representation itself. defined in ch. 1 — open in glossary , which is a kind of machine learning machine learning Acquiring knowledge by extracting patterns from raw data instead of hard-coding it. defined in ch. 1 — open in glossary , which is one approach (of several) to AI. Each ring below carries an example technology:

The placement matters: a knowledge base knowledge base approach Hard-coding world knowledge in formal languages for automatic logical inference (e.g. Cyc). defined in ch. 1 — open in glossary is AI without learning; logistic regression learns but consumes hand-made features; a shallow autoencoder autoencoder An encoder plus decoder trained to preserve information while giving the representation useful properties. defined in ch. 1 — open in glossary learns its representation but only one step of it; an MLP multilayer perceptron A function composed of many simpler functions; the quintessential deep model (feedforward deep network). defined in ch. 1 — open in glossary learns representations of representations.

What each discipline learns (fig 1.5)

The same story as a set of flowcharts. Shaded boxes are the components able to learn from data — watch the shading climb as you move right:

§1.1 Who should read this book?

The book targets two audiences: university students (undergraduate or graduate) entering machine learning or starting deep-learning/AI research, and software engineers without an ML/statistics background who want to acquire one quickly and use deep learning in a product — a field already proven useful in computer vision, speech and audio, NLP, robotics, bioinformatics and chemistry, video games, search, online advertising and finance.

The three-part organization serves both audiences, and you are encouraged to skip what you already know:

• Part I — applied math and ML fundamentals. Skip it if you know linear algebra, probability and basic ML.
• Part II — the established, essentially solved deep-learning technologies. Stop here if you just want to build working systems.
• Part III — the more speculative research ideas widely believed to matter for the field’s future.

One assumption is made of everyone: a computer-science background — programming, a basic grasp of computational performance and complexity theory, introductory calculus, and some graph-theory terminology.

The book’s own roadmap (fig 1.6) shows which chapters are prerequisites for which — an arrow means “read the tail before the head”:

Note what the arrows say: within Part II, everything flows from chapter 6 — regularization, optimization, CNNs and RNNs are four parallel elaborations of the feedforward network, and all four feed the two practice-facing chapters (11, 12). Part III fans out again: chapters 13–15 extend the representation-learning story, while 16 → 17 → 18 → 19 build the probabilistic machinery that chapter 20 assembles into deep generative models.

§1.2 Historical trends — the four claims

The book closes its introduction by promising a history told through four key trends, each expanded in the coming sections:

1. Many names, waxing and waning: deep learning has a long history under different philosophical banners — its popularity has risen and fallen twice before (§1.2.1).
2. More data → more useful: deep learning becomes more valuable as training data grows (§1.2.2).
3. Bigger models over time: hardware and software infrastructure keep allowing larger networks (§1.2.3).
4. Harder tasks, higher accuracy: deep learning solves increasingly complicated applications with increasing precision (§1.2.4).