Model Fusion: a New Approach to Processing Heterogenous Data

In many practical situations, it is necessary to extract information from data of different types. For example, in geosciences, several sources of data can be used to determine the interior structure of the earth: the first arriving waves from earthquakes and from manmade sources, measurements of the earth’s gravity field, measurements of the dispersion of surface waves generated by earthquakes, etc. At present, most existing data processing techniques deal only with data of one type. A joint use of all the information derived from multiple types of data sources is an important theoretical and practical challenge. Such a joint use would provide the best description of the object of study; for example, in geosciences, such a joint “inversion” would represents the best model of the interior structure of the earth. While such combination methods are being developed, as a first step, this thesis proposes a practical solution: to fuse the models generated from different datasets by the existing data processing techniques. In geosciences and in other applications, models generated from different datasets not only have different accuracy and coverage, but also different spatial resolution. This thesis describes how to optimally fuse such models under interval and probabilistic uncertainty. Additional ideas are described that take into into consideration the inherent discrete vs. continuous nature of different models, and how to gauge accuracy of each of the fused models. The resulting techniques can be used in various application domains to merge models of different accuracy, coverage, and spatial resolution.