دانلود کتاب یادگیری ماشین در مدل سازی و شبیه سازی

Machine learning (ML) techniques have recently become remarkably successful. Although some of the basic principles were established decades ago, the advent of GPU-enabled numerical analysis and parallel computing, coupled with some recent theoretical advances, has led to significant ML progress in diverse fields, such as image and speech recognition, natural language processing, bioinformatics, and game theory. Many of the developed applications require large amounts of data to be used as training sets. However, also, tremendous advances have been made in having a computer algorithm learn the problem solution from a reasonable amount of data and a simple set of rules.
While ML approaches have been applied in a number of areas such as economics, e-commerce, image recognition, language interpretation, and medicine, their full potential in “Modeling and Simulation” still needs to be explored. Indeed, ML approaches have the potential to drastically reduce the computational cost—from design to analysis back to design—and allow challenging problems in engineering and the computational sciences to be solved which have never been solved before, in fact not even been tackled before.
Machine learning algorithms can be separated into two main classes: supervised and unsupervised learning. In both classes, the algorithms have access to a set of observations for the role of training data. However, the nature of the training data, and hence what can be extracted from them, differs. In supervised learning, the training data consists of a given set of input values and a given corresponding set of output values. Using the training data, the ML algorithm tries to find the best parameters of a prediction function that can accurately emulate the output as a function of the input, each of interest. On the other hand, in unsupervised learning the output of interest is not known a priori and an effective use of the data is much more challenging. The third intermediate class exists, semi-supervised learning; in this case, some, but not all of the given input values have given corresponding output values.