Название: Machine Learning Control by Symbolic Regression
Автор: Askhat Diveev
Издательство: Springer
Год: 2021
Страниц: 162
Язык: английский
Формат: pdf (true), epub
Размер: 14.2 MB

This book provides comprehensive coverage on a new direction in computational mathematics research: automatic search for formulas. Formulas must be sought in all areas of science and life: these are the laws of the universe, the macro and micro world, fundamental physics, engineering, weather and natural disasters forecasting; the search for new laws in economics, politics, sociology. Accumulating many years of experience in the development and application of numerical methods of symbolic regression to solving control problems, the authors offer new possibilities not only in the field of control automation, but also in the design of completely different optimal structures in many fields.

For specialists in the field of control, Machine Learning Control by Symbolic Regression opens up a new promising direction of research and acquaints scientists with the methods of automatic construction of control systems. For specialists in the field of machine learning, the book opens up a new, much broader direction than neural networks: methods of symbolic regression. This book makes it easy to master this new area in machine learning and apply this approach everywhere neural networks are used. For mathematicians, the book opens up a new approach to the construction of numerical methods for obtaining analytical solutions to unsolvable problems; for example, numerical analytical solutions of algebraic equations, differential equations, non-trivial integrals, etc.
For specialists in the field of artificial intelligence, the book offers a machine way to solve problems, framed in the form of analytical relationships.

Machine Learning is one of the areas of artificial intelligence associated with solving problems based on algorithms that can learn or gradually improve the performance of a given task. Machine learning is based on the idea that computing systems are able to show a behavior that was not explicitly programmed in them, they can identify patterns, rules, or functional dependencies and make decisions on their own. Control systems can act as such functional dependencies. In machine learning, the control system is learned, not programmed by the developer.

The most famous Machine Learning technique now is neural networks, and sometimes Machine Learning is equated with neural network training. This is incorrect, because machine learning is a broader concept. It includes such early forms of data analysis as probabilistic modeling based on the application of Bayes’ theorem and logistic regression; classification algorithms such as kernel methods, hierarchical structures such as decision trees, random forests, and gradient boosting. We definitely note that deep neural networks show the best performance in many tasks, which explains their popularity. At its core, a neural network is a function with a specific structure and a large number of unknown parameters. Learning, or, more precisely, training neural networks is finding optimal values of parameters.