×

Novel learning tasks, optimization, and their application. (English) Zbl 1102.68543

Schwefel, Hans-Paul (ed.) et al., Advances in computational intelligence. Theory and practice. Berlin: Springer (ISBN 3-540-43269-8/hbk). Natural Computing Series, 245-318 (2003).
Summary: This chapter describes methods for learning and optimizinig solutions to engineering problems. Where standard learning tasks do not fit the requirements of the applications, they are varied. In particular, the use of prior knowledge and unlabeled examples may ease a learning task. In contrast, learning target concepts that change over time and constructing new features are challenging learning tasks. The changes of the learning tasks that ease as well as those that challenge learning cover a broad range of application problems. Solutions to the new, nonstandard learning tasks are developed, integrating numerical optimization strategies, evolutionary algorithms, support vector machines, neural networks, and fuzzy controllers. A workbench for learning methods allows their systematic evaluation with respect to the engineering problems.
The chapter is structured as follows: First, extensions of standard learning tasks are defined, and methods for their solution are proposed and evaluated. Second, a problem in chemical engineering is presented in Section 8.2, namely the determination of suitable models for batch chromatography. This process is a standard separation method in the pharmaceutical industry. The critical model parameters of physical and grey-box models of the process are determined from measured concentration profiles. Mathematical parameter estimation and neural networks are used for this purpose. Third, from the field of electrical engineering, circuit design, device modeling, and quality assessment tasks are covered in Section 8.3. An evolutionary algorithm for the optimization of electronic circuit components is presented. Furthermore the construction of simulation models for electronic device development and circuit control is investigated. Different algorithms and hardware implementations of neural networks for microelectronics and process control are examined.
For the entire collection see [Zbl 1001.00010].

MSC:

68T05 Learning and adaptive systems in artificial intelligence

Software:

DistAl; VODE; JiffyTune
PDFBibTeX XMLCite