|Practical Machine Learning|
Page 3 of 3
Author: Sunila Gollapudi
Chapter 11 Deep learning
This chapter relates to neural networks, which aim to learn and find solutions using input data. The chapter opens with an overview of what deep learning is, before looking at neural networks with reference to architecture and the neuron model. A simple digit recognition example is provided. Next, various taxonomies are described, including: convolution neural networks, deep Boltzmann machines. The chapter ends with links to code that implements artificial neural network and related algorithms using each of the 5 tools. I found this chapter interesting but difficult to follow, perhaps more detailed explanations are needed.
Chapter 12 Reinforcement learning
Reinforcement learning is different from the traditional supervised and unsupervised learning techniques, instead it learns using feedback from the environment (which produces a new situation), and it is both iterative and adaptive. The chapter opens with a brief review of supervised and unsupervised learning, providing a context for reinforcement learning. The Markov Decision Process, which is key in understanding reinforcement learning, is discussed, and a basic agent-environment model is examined with reference to rewards and punishments.
Next, various methods to solve reinforcement learning problems are briefly examined, including: Dynamic Programming, Temporal difference learning, and Q-Learning. Most of these descriptions are quite terse, so be prepared to look elsewhere for detail. The chapter ends with links to code that implements artificial neural network and related algorithms using each of the 5 tools.
Chapter 13 Ensemble learning
This chapter is concerned with combining results to obtain more accurate results, allowing better business decisions to be made. The chapter opens with a look at the concept of “The wisdom of the crowds”, a method where the results from combined analysis are often better than from individuals (e.g. the crowd answer in” Who Wants To Be A Millionaire” is often correct!).
The chapter continues with a discussion of ensemble learning method use cases, including recommendation systems, and anomaly detection. The section ends with a look at various supervised and unsupervised ensemble methods – accompanied with a fair amount of math. The chapter ends with links to code that implement supervised ensemble learning algorithms using each of the 5 tools.
Chapter 14 New generation data architectures for Machine learning
This chapter discusses some advanced and upcoming Machine Learning architectures and technologies. The chapter opens with a brief look at the traditional architectures, and the impact of Big Data on the creation of newer distributed parallel processing architectures is examined.
The chapter then looks at some emerging architectures for Machine Learning, including: semantic data architecture, multi-model database architecture / polyglot persistence, and Lambda architecture. In each case, the architecture is outlined, and a list of vendors given.
This book aims to introduce you to both basic and more advanced features of Machine Learning. The book is generally well written. There are helpful diagrams, and inter-chapter links. Each chapter ends with a helpful summary and relevant example code (in R, Julia, Python Mahout, and Spark). An understanding of math and generally programming is required, since little explanation is given.
The book is certainly very wide in its scope, perhaps to the detriment of its depth, often providing just a short overview of a topic. Some sections have so little detail, it makes me wonder if they are worthwhile including (e.g. chapter 12’s methods to solve reinforcement learning problems). I do wonder if putting Machine Learning in the context of Big Data and Hadoop is necessary, perhaps the book should focus on one thing - Machine Learning!
The book should have included walkthroughs of the example code rather than delegate it out-of-scope, that said, the plentiful code is generally well-commented.
Overall, the book provides a useful, wide ranging review, of the current use of Machine Learning.
|Last Updated ( Saturday, 28 November 2020 )|