Computing statistics under interval and fuzzy uncertainty. Applications to computer science and engineering (Q653914)

This book is a research exposition by Kreinovich and coworkers. Embedded into a general terminology it contains mainly chapters which go back to papers of this research group published during the last ten years. The main goal is to present algorithms for computation of statistical characteristics (like variance) but under interval and fuzzy uncertainty of the available data. In this book, fuzzy uncertainty is reduced to interval uncertainty by alpha-cutwise consideration of (convex) fuzzy uncertainty. In many cases, the general algorithm for an assigned problem appears as NP-hard. Then the authors look for natural restrictions (e.g. for the uncertainty intervals) which allow to present feasible algorithms with lower (e.g. polynomial) complexity. Although this is not a textbook the presentation is more or less selfcontained, i.e. for the reader previous knowledge is of advantage but not necessary. For increase of readability, mathematical proofs are presented always at the end of the chapters. Brief summary of the contents: Part I (60 pages) presents an overview of techniques which will be used lateron, e.g. an introduction to fuzzy terminology, interval mathematics (looking for ``good'' enclosures), analytical means (e.g. linearization), complexity of algorithms, and decision theory. Part II (170 pages) as core of the book presents algorithms (for computation of mean, median, variance, higher moments, outlier thresholds, covariance and correlation, entropy) and discusses and improves their complexity. Possibilities of parallelization and problems of accuracy are also discussed. Part III (15 pages) is a short discussion on gauging the quality of the input data. Part IV (80 pages) presents applications in bioinformatics (cancer research), in computer science (optimal scheduling of supercomputing e.g. in looking for patterns in long DNA or RNA sequences), in information management (estimation of the degree of trust in an agent), in signal processing (radar observation), in computer engineering (time analysis of computer chips), in mechanical engineering (failure analysis of materials) and in geophysics (seismic inverse problems). Finally, Part V (60 pages) looks beyond interval and fuzzy uncertainty and takes e.g. constraints into account, deals with discontinuous processes, with smooth distributions, with interval bounds on the probability density function, with interval-valued and type-2 fuzzy techniques.