The Hundred-Page Machine Learning Book by Andriy Burkov — Summary & Review

Burkov’s book is a no-nonsense, brisk walkthrough of the core ideas behind machine learning (ML), packed into about 140 pages. The promise is bold: give you a solid grasp of the essentials without drowning you in details or hype. And it mostly delivers.

The book kicks off by breaking down the two main flavors of ML: supervised and unsupervised learning. Think of supervised learning as the student who’s got a teacher showing the way—labels and outcomes guide the model’s learning. Unsupervised learning, by contrast, is more like wandering in a forest without a map, trying to find patterns and groups in data without any explicit instructions. This basic split is crucial because it sets the stage for everything else.

From there, Burkov drills into some of the workhorses of ML: Support Vector Machines (SVMs), neural networks, and ensemble methods. SVMs get a neat explanation as a way to draw the best possible line or boundary between categories in data. Neural networks, which often get hyped to the moon and back, are introduced with enough math to understand what’s happening under the hood, but not so much that you feel like you’re back in calculus class. Ensemble methods get their due as clever ways to combine multiple models to squeeze out better predictions—because sometimes, one brain just isn’t enough.

The book doesn’t shy away from the gritty details either. Gradient descent—the optimization algorithm that’s the backbone of training many models—gets a clear treatment, showing how models learn by iteratively tweaking parameters to minimize error. The chapters on clustering and dimensionality reduction explain how to group data and cut down on noise or irrelevant features, which is crucial when dealing with messy real-world data.

Burkov also touches on some more advanced but still approachable topics like autoencoders and transfer learning. Autoencoders are a neat trick for unsupervised learning, compressing data and then reconstructing it to find hidden structures. Transfer learning, meanwhile, is the idea of borrowing knowledge from one problem to tackle another—a concept that’s become a staple in modern ML.

What’s refreshingly practical is the attention given to feature engineering and hyperparameter tuning. These are the behind-the-scenes crafts that often decide whether a model is just okay or actually useful. Burkov doesn’t pretend these are glamorous, but he makes it clear they’re essential.

The book is peppered with mathematical formulas and intuitive illustrations. It doesn’t shy away from math—if you’re squeamish about equations, this might feel like a cold shower. But the math isn’t gratuitous; it’s there to clarify, not confuse. The examples ground the theory in practical contexts, though you won’t find hands-on coding exercises or projects here. This isn’t a cookbook for ML practitioners who want to get their hands dirty; it’s more like a clear, compact reference manual.

That said, the book assumes you’re not a total novice. You should be comfortable with basic math concepts and some notation. If you’re starting from zero, you might find yourself pausing, Googling, or wishing for more hand-holding. Also, the book doesn’t dive deep into the latest ML crazes like deep learning architectures, natural language processing, or reinforcement learning. It’s a solid foundation, but if you want to keep pace with the bleeding edge, you’ll need to look elsewhere.

In a field that’s flooded with hype, jargon, and endless tutorials, Burkov’s book is a breath of fresh air. It’s concise without being dumbed down, practical without being simplistic. It’s the kind of book you can read in a weekend and walk away with a coherent mental map of what ML actually is—and what it isn’t. For anyone who’s tired of buzzwords and wants a clear-eyed introduction, this is a good place to start.