Cambridge University Press Understanding Machine Learning: From Theory to Algorithms

Livro muito bem escrito que envolve da teoria a prática em Machine Learning. Esse livro é um dos melhores na área de ciência de dados.

Ich bin Informationstechnik Student und wollte ein bisschen mathematischen Background haben. Zu dem Buch gabs auch eine Lehrveranstaltung an unser Uni. An sich ist das sehr gut, einige Kapitel verstehe ich auch sehr gut und auf anhib, andere überhauptnicht. Wenn ich mehr Zeit investieren würde wären die sicherlich auch kein Problem, aber bei manchen Kapiteln sehe ich den Mehrwert nicht. Jedoch ist das Buch eine gute Einführung ins Maschine Learning, um auch ein wenig Background Informationen zu bekommen.

Taking the Shai's course this term. A fantastic book as supplementary material. There are some typos but not influencing the whole understanding.

I have read many of the main books on machine learning. This is hands down the best. Rather than a laundry list of techniques, the book starts with a concise and clear introduction to statistical machine learning and then consistently connects those concepts to the main ML algorithms. Each chapter is 10 pages or so of crisp math and lean prose. A brief summary at the beginning of each chapter gives a clear sense of what will be accomplished in it, and attention to notation makes sure that mathematics supports understanding rather than getting in the way. This is definitely not a "how to" book, but rather a "what and why" book, focused on understanding principles and connections between them. I read the book cover to cover, and I was left with a sense of machine learning as a coherent discipline, and a solid feel for the main concepts.

Simply the best book on theoretical machine learning I have come across. The book does not only provide, very well written, rigorous proofs but an exceptionnel coherent and insightful presentation of the whole subject of learning. The point of view is essentially distribution agnostic, that is it emphasize the discriminant approach of PAC learning and notions of complexity (VC and Rademacher) and does not go into details of any probabilistic models (as does another excellent book by Ch. Bishop). This book definitely won't help you win any Kaggle competion but will make you very happy if you care to go to the bottom of things.