Browse Books

Reviewer: Joseph M. Arul

The current research in parallel and distributed machine learning for large datasets is presented in this book. The applications discussed are mainly in the financial and petroleum industries. The preface states: "The book will be useful to the broad audience of researchers, practitioners, and anyone who wants to grasp the future of machine learning." However, this book will be more useful to current researchers in the field and those who are familiar with the techniques, rather than beginners. It is a collection of contemporary studies, which can enhance the knowledge of those already in this field who are well established in parallel and distributed machine learning. It is highly technical. Various scholars present contemporary topics in machine learning and scaling up, to motivate researchers to explore further. An in-depth exploration of techniques such as supervised and unsupervised algorithms is included for anyone interested in delving deep into the field of machine learning. Recent evolution in hardware architectures and programming frameworks has made it convenient to exploit the parallelism possible in many learning algorithms. In many modern applications, large datasets are often accumulated on distributed storage platforms, which further motivates the development of learning algorithms in a parallel or distributed environment for existing sequential algorithms. Anyone aiming to achieve speedup, efficiency, the scaling up of their algorithms, and performance gain will find this textbook very useful. The book is divided into four parts. The first part describes frameworks for scaling up machine learning, mostly using the k -means algorithm, which can be used in various applications such as decision tree ensembles, frequent pattern mining, and regression. Currently used frameworks for such scaling machines are MapReduce, DryadLINQ, the message passing interface (MPI), and CUDA. The second part of the book deals with supervised and unsupervised learning algorithms. The supervised learning algorithms utilize the training of data to construct a prediction function f ; thus, f is applied to test instances. In the unsupervised learning algorithm, the data is clustered to construct a function f that partitions an unlabeled dataset into k =| Y | clusters, with Y being the set of cluster indices. It deals with parallelizing support vector machines (SVMs), boosted decision trees, belief propagation (BP), and a Markov chain Monte Carlo (MCMC) technique. Part 3 presents the traditional supervised and unsupervised learning formulations, focusing on parallelizing online, semi-supervised, and transfer learning. The final part presents several learning applications in distinct domains with the main focus on scaling up, which is crucial to computational efficiency and improving accuracy. This book explains in detail the computing platforms, learning algorithms, prediction problems, and application domains for a variety of parallelization techniques for scaling up machine learning. It is not for anyone trying to understand the basic concepts of parallelization and distributed environments, but could be an excellent resource for researchers in the field. Online Computing Reviews Service