In prioritizing geometric intuition, algorithmic thinking, and applications our treatment of machine learning is more intuitive, functional, and fun than what is currently available to readers. Some of the unique pedagogical, topical, and technical features of our book include:
1. A presentation built on lucid geometric intuition
We use a continuous geometric paradigm throughout the text, including over 200 color illustrations, that gives readers a deeper understanding of concepts as well as the connections between traditionally disparate topics.
2. A learning experience where students learn by doing
We reinforce readers' understanding of machine learning concepts by encouraging them to code up what they study in the body of the text using practical examples and applications. To this end we provide over 50 in depth coding exercises drawing on a diverse set of synthetic and real data sets.
3. Inclusion of real application to motivate concepts
We provide a breadth of applications throughout our book, taken from disciplines including computer vision, natural language processing, economics, neuroscience, recommender systems, physics, biology, and more.
4. A rigorous yet user friendly presentation of state-of-the-art numerical techniques
Not only do we provide optimization schemes for every model presented in the book, but dedicate space to completely describing the most cutting edge optimization tools used in practice today.
5. Inclusion of feature design and learning as major topics
Because features play a prominent role in terms of performance of learning algorithms we give feature design and learning prominent focus throughout our book.
6. A fused introduction to logistic regression and support vector machines
We present both of these popular classification methods in the same geometric framework, helping readers understand why they perform so similarly in practice. We also provide a similarly unified description of One versus All and softmax regression multiclass classifiers.
7. An unparalleled treatment of advanced topics through the lens of function approximation
By employing the perspective of function approximation we provide a wholly original and illuminating presentation of feature learning methods, cross validation, neural networks, and kernel methods.
8. A refined introduction to deep neural networks and kernel methods
Neural networks and kernel methods are introduced jointly, providing a smoother comparison of these two popular tools, as well as a shared platform on which to discuss their respective strengths and weaknesses.
9. An elucidated description of kernels and the array of kernelizable models
By employing a flexible approach to kernelization we help the reader to appreciate how a wide array of machine learning models are kernelizable including: SVMs, linear/logistic/softmax regression, K-means, PCA, etc.