Assessing the Performance of Data Fusion Algorithms Using Human Response Models

There is ongoing interest in designing data fusion systems that make use of human opinions (i.e.,"soft" data) alongside readings from various sensors that use mechanical, electromagnetic, optical, and acoustic transducers (i.e., "hard" data). One of the major challenges in the development of these hard/soft fusion systems is to determine accurately and flexibly the impact of human responses on the performance of the fusion operator. Examples, counterexamples, and thought experiments have been used to illustrate specific performance aspects of soft and hard/soft fusion operators. However, these methods do not provide a systematic means for calculating performance statistics. Data sets of human responses developed through human testing have also been used. However, results obtained in this manner were often hard to generalize and difficult to tune up due to the experimental and administrative limitations imposed on human testing. Models of human decision-making and confidence assessment from cognitive psychology offer a unique opportunity to assess the performance of fusion systems which make use of human opinions. Such models, which can programmed and modified readily and inexpensively, can be used to assess the performance of hard/soft fusion systems and to make credit assignments to the multiple sources of data that lead to the final estimate/decision of the fusion architecture.;The main contribution of this thesis is a set of algorithms for determining the performance of various soft and hard/soft fusion operators using existing models of human decision-making. In the first part of the thesis, we discuss the current state of the art and introduce applicable techniques for hard/soft fusion. In the second part of the thesis, we present techniques and examples for determining the performance of a family of soft and hard/soft fusion operators on a set of binary and multihypothesis tasks. We also present a method for simulating "vague" human responses, and correspondingly assess the statistical performance of soft fusion systems that have been proposed for combining sources of such imprecise information.