Identification, Adaptation, Learning: The Science of Learning Models from Data

This book offers a tutorial view of recent trends in the science of modelling, adaptation, and learning. The most important modern approaches to identification, namely the shastic, behavioral, subspace, and frequency domain approaches, are discussed thoroughly. On adaptation, tuning the parameters of a linear model is presented as a cure for uncertainty, and the asymptotics of recursive least squares and self-tuning systems are explained simply after a fully deterministic analysis. For constructing nonlinear models from data, neural networks and wavelets are considered as useful nonlinear tools, and fuzzy logic is presented as a way of coping with qualitative information. A final chapter deals with optimization methods. The book will become an important reference for researchers in the field.