By Sergios Theodoridis.
If you are very comfortable with math and want to cover a wide range of Pattern Recognition and Machine Learning techniques, this is the book for you. For everyone else it will be a chore to read through. Unfortunately the book is bogged down with heavy math notation. For the more advanced algorithms, it is next to impossible to get a feel of what the algorithms are and how they work if you are not already familiar with them. Instead of algorithm design, you get the math representation of what the algorithm tries to acheive, then some proof about optimality, then some proof that nobody asked for, and then some more math notation.
Even if one can take extreme math notation to the chest and keep going, they would still find the book a chore. The first 3-5 chapters are overly verbose, where the author goes on and on about the peripherals without touching on the meat of the subject nearly as much. Also, for a lot of the algorithms in these chapters (particularly in the Linear Classifiers chapter), the author opted to go for the counter-intuitive and inefficient solution simply because the math behind it is easier. I understand that opting to go for an easier route to juggle notation is a sound thing to do, but you are sacrificing usefullness.
Some of the chapters in the book are clearly rushed, like the Context-Dependent Classification chapter and a couple of the Clustering chapters. Where, by the way, the author again goes for verbose description instead of brevity.
Now on the positives. The book is pretty good as a reference, if you know an algorithm and want to refresh your memory on the details.
That’s pretty much it. I don’t recommend it, this was pretty much a waste of my time.