Deep Learning (Adaptive Computation and Machine Learning)

Great (and a very timely & relevant) book on this exciting and cutting edge domain. A much needed reference material on Deep Learning, covering the foundations, as well as offering a glimpse into the further research areas.

This book covers both basic and advanced topics. It not only tells you 'what' but also explains 'why'. The author makes many obscured concepts very easy to understand. The quality of the book is also excellent. Highly recommend!

Superficial. It is a book that is neither very technical nor very practical. It is not enough to begin to understand the networks from an operational point of view, nor does it provide enough formalism to understand them from a theoretical point of view. The book is a kind of literature review and tries to cover many concepts that require more elaboration in order to really understand them.

It is known that deep learning has no solid mathematical foundation. This book makes me believe it more. The authors talked too much on the methods of traditional statistical machine learning since there is little valuable mathematics of deep learning. You may find most of its content in any textbook of machine learning.

Good book, it is a good intro for beginners. Basic concepts are explained in the book so it is self-contained and covers the deep learning field. If you are an expert, then another books may better, but if you want to learn about Deep Learning, I recommend it.

This is the best book that introduces the field of Deep Learning from theoretical & practical prospectives. The book lays the foundation in the first two parts and put you up to date in terms of research in the last part. Truly the right book for DL.

This book has by far surpassed my expectations! I have purchased many machine learning and deep neural network books in the past, but nothing has ever come close to this book! First of all, it is written by the fathers of Deep Learning, and is therefore an authority. Secondly, the book is broken into three parts: 1. A math overview and refresher. 2. Deep Learning applications and 3. Research in Deep Learning. I can't help but go through this book from front to back. It is a smooth read, and every sentence written is meaningful. These guys know their stuff! And after you read this book, YOU WILL ALSO know your stuff!
If you feel daunted by the price, just remember, you get what you pay for! I'd say they could easily charge about $300+ for this book, but they are doing everyone a very kind favor by ONLY charging this reasonable amount. You get A LOT of bang for your buck with this purchase. I hesitated at first about buying this book because of the price, but I am soooooo happy that I did! Worth every penny! Look no further, get this book and start your Deep Learning journey!!

This book is a valuable addition to my machine learning library. It's mostly easy to read given enough statistical maturity, say at the level of All of Statistics: A Concise Course in Statistical Inference (Springer Texts in Statistics). Definitely easier than The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition (Springer Series in Statistics)

I like the fact that they use clear logical arguments and reference papers to justify certain choices like the RELU unit. I remember reading the classical approximation theorems by Hornik and getting an impression that the unit has to be bounded for universal approximation, which RELU is certainly not, but apparently it has good approximation properties too. I am also glad that they discuss issues of regularization and not just blindly assume that deeper is better as some literature and hype seems to suggest.

About 1/3 of the book is background in relevant machine learning basics and optimization, mostly part I. Convex optimization is very much an active research field now, and they describe some new interesting algorithms. The rest of the book surveys various deep learning architectures. Based on the organization of the material, I would guess they consider convolutional and recurrent networks to be the main architectures that should be used in certain specific application. I'm surprised that they consider Boltzmann machines to be research rather then current practice, but they are the experts.

What is missing is a clear indication of a good deep learning architecture to compete generally with black-box methods like random forest. I have looked at a number of other, essentially useless books that claim to be easy, but just end up using various libraries and make arbitrary architecture selections based on some paper. Even books with concise implementations of most major learning algorithms, like Commodity Algorithms and Data Structures in C++: Simple and Useful, don't describe a clean black-box deep learning implementation. I am hoping that upcoming books like the next edition of Data Mining: Practical Machine Learning Tools and Techniques, Third Edition (Morgan Kaufmann Series in Data Management Systems), or deep learning specific Patterson-Gibson or Buduma book (the store won't let me link future books) will describe a good black-box option.

On the funny side, I find it interesting that a supposedly busy-busy CEO Elon Musk has taken time to read and endorse this. So far I have only seen Bill Gates endorse Knuth books.

This was a very disappointing book. It was not good at presenting a conceptual framework, a gist if you will. But neither did it go into details of the technology. Instead, it exhibited a needless fascination with linear algebra and a sophomoric obsession with equations. The purpose of this book was not to impart information, but to present as much math as possible... almost as if this was to satisfy a school assignment--do a cursory review of the field, and write down as many equations as you possibly can... no algorithms, no useful details, just lots and lots of equations. Frankly, I cannot tell who the intended audience is, but it is definitely not me. As an aside, I have seen this facade of mathematical rigor before, and it is usually because the foundations are shaky (like back when they thought you could make intelligent machines only if did could perform more and more logical inferences, predicate calculus, or some such.) Methinks I will pass on mindless learning for now.

This book was poorly written, too difficult for people who want to ramp up with DL. Even with a PhD degree in EE and previous background in ML related DSP, I felt the content in this book loosely organized and hard to digest. Many formulas lack deduction, and results were simply listed without proofs. It reads more like a compilation of tech reviews.

The first chapters provide a great review of the underlying fields of mathematics, with important motivational comments about the relationships between general purpose optimization and deep learning. The basics of deep learning are presented at a fine level and with exceptional clarity. The book is very well structured and authored, the writing style is really exceptionally clear relative to the topic.

As the field is still rapidly developing, one would probably want to catch up on advances made following the book's time of writing, but either way it lays down good foundations.

I'd imagine reading this book as a preview for walking through a tutorial of one of the deep learning frameworks out there. Last but not least, the text is freely and readily available in good html format online.

A good breadth of coverage. Nevertheless, I am expecting a rigorous textbook, not brief rephrases of literature, which I constantly feel while reading the second half of the book. In this sense, this may not be a proper introductory TEXTBOOK for new-comers of the area. I have some backgrounds in Statistics and Information Theory, and I do not like the treatment of those fundamentals in the first few chapters, which were not well-organised.

In general, this is more a compilation of lectures or talks, not rigorous text to me.

I would still recommend Bishop(2006), though it not focuses as exclusively as this one on deep networks.

I have mixed feelings about this book. For a researcher, it's a great reference written by researchers who obviously know their stuff. For an instructor, it failed as a graduate-level textbook. A book like this isn't suited or accessible to bridge the gap for someone who has taken a linear algebra, calculus, and prob/inf theory course to apply that knowledge to neural networks. Fortunately, as time goes by, new resources are added to the book's website, but some of those came too late for me during the first semester that I adopted the book.

Although authors are pioneers in the field and did great work, the book is very poorly written. Too much abstract information, lots of unnecessary details and lots of missing information. It was really frustrating experience reading this book.

It's definitely THE authoritative reference on Deep Learning but you should not be allergic to maths. That said reinforcement learning is superficially exposed which is due for an additional chapter [Note].

The main weakness of this masterpiece is the lack of practical programming exercices left to a companion web site. But to cover all the practical stuff, the book should have exceeded 775 pages that it already has.

I dream of he same content in the form of a series of iPython Notebooks with all exercices and code samples using Keras, TensorFlow and Theano.

[Note] To be completely honest the authors wrote a short disclaimer in the «Machine Learning Basics» chapter 5, page 103 about reinforcement learning. « Such algorithms are beyond the scope of this book ».

Despite its title, this book provides a superb introduction to much of machine learning. Similarly, its treatment of Deep Learning is easily one of the best. The care with which the authors wrote is obvious, starting with systematically defining terms, a mark of true scholarship in my opinion. I can't over-recommend this book. And of course, the authors are all deeply involved (no pun intended) in developing Deep Learning in the first place. It doesn't get any better.

On the accompanying web-site, there is an easy to print "HTML" version of the book. They don't have PDF version, because of the contract with MIT that forbids it, but what they have to print is even better. It's free, obviously.

Strongly theoretical, i'd like to have code examples of the pseudocode presented in the book, however it's what i expected, the clearest mathematical definition of neural networks, training methods etc... highly recommendable