» Back to Press Release
Media Release
Media Contact:
Amanda Yun
Senior Editor / Press Officer
Tel: (65) 64665775, Ext. 446
Email: heyun@wspc.com
Data Fusion: How to Make Better Decisions with Incomplete Information
As the world is awash with data obtained from numerous and varied processes, there is a need for appropriate statistical methods which in general produce improved inference by taking as input the information from many sources. There is no single definition of "data fusion", but regardless of one's view, the underlying idea is to come up with estimates or decisions based on multiple data sources as opposed to more narrowly defined estimates or decisions based on single samples. Real data can be fused with other real data, or even with artificial "fake" data. Thus, a given sample can be fused with computer-generated data giving rise to the notions of "out of sample fusion" and "repeated out of sample fusion". Statistical Data Fusion by Benjamin Kedem, Victor De Oliveira, and Michael Sverchkov, deals with some aspects of statistical inference based on the combination, or fusion, or integration of many samples, each obtained from (in general) an unknown probability distribution. An important example of data fusion where the idea is to "borrow strength" from different samples, is the important problem of small area estimation in the context of survey sampling.
In a recent article on science and technology in The Economist (February 4, 2017, pp. 67-69) titled "Better than real", The Economist tells readers that if high tech giants such as Google and Microsoft "have their way, then the next thing to leap from fiction to fact will be augmented reality (AR)." The article continues by saying that AR supplements the real world by laying over it "useful or entertaining computer-generated data." This is precisely the message in Statistical Data Fusion. A specific case in point is the estimation of very small exceedance or threshold probabilities, where the problem is to estimate the small chance that quantities such as mercury in fish, an insurance claim, blood pressure, and the plutonium level next to a nuclear reactor, exceed high alarming levels. Thus, suppose the real data consist of mercury levels obtained from 100 fish in a certain region of the ocean, and suppose that the 100 mercury levels are entirely safe. Is it then possible to come up with inference about the small chance of mercury exceeding an unsafe level when in all likelihood the data indicate otherwise? Apparently the data at hand are insufficient for answering this question. Not so fast, however; in the spirit of augmented reality, it is possible to fuse the real data repeatedly with certain computer generated random numbers to obtain reliable interval estimates containing the small chance in question. This fact has a theoretical basis supported by numerous computer simulations.
Statistical Data Fusion has been written with the data practitioner in mind. It offers examples and illustrations of a wide range of possible applications: from a novel approach, to time series prediction, to the estimation of small tail probabilities, to the estimation of mean body mass index in United States counties.
This book retails for US$98 / £81, and is available on Amazon, Barnes and Noble, and other major online booksellers. To know more about the book visit https://www.worldscientific.com/worldscibooks/10.1142/10282.
About the Authors
Benjamin Kedem, who joined the University of Maryland in 1975 as an Assistant Professor, holds a PhD. (1973) in statistics from Carnegie-Mellon University, Pittsburgh, and specializes in inference for space-time stochastic models. His research has been recognized by several awards including IEEE W.R.G. Baker award (1988), award from the Armament Development Authority, Israel (1984), NASA/Goddard Exceptional Achievement Award (1997), and IBM Faculty Award (2006). He is a Fellow of the American Statistical Association.
Victor De Oliveira is a Professor in the Department of Management Science and Statistics at the University of Texas at San Antonio College of Business. As a statistician, De Olivera focuses primarily on Bayesian Methods, Environmental Statistics, Geostatistics, Markov Random Fields, Spatial Prediction and Space-Time Modeling. In addition to this, De Oliveira has contributed to the creation of BTG software, which performs Bayesian prediction of transformed Gaussian random fields, and the R package geoCount, which performs analysis and modeling for geostatistical count data.
Michael Sverchkov is a research mathematical statistician at the Bureau of Labor Statistics. His areas of research are estimation and inference with complex sample survey data, variance estimation, small area estimation, calibration, time series analysis.
About World Scientific Publishing Co.
World Scientific Publishing is a leading independent publisher of books and journals for the scholarly, research, professional and educational communities. The company publishes about 600 books annually and about 140 journals in various fields. World Scientific collaborates with prestigious organisations like the Nobel Foundation and US National Academies Press to bring high quality academic and professional content to researchers and academics worldwide. To find out more about World Scientific, please visit www.worldscientific.com.