Abstract:
|
Data fusion is the combination of disparate data from different modalities to answer a single question of interest. Each piece of data provides a window into the inference or prediction problem where no single data stream can provide the whole picture. This paradigm is most beneficial to situations where a large amount of data can be collected to inform a small number of observations, i.e., big data with small n. Combining diverse data or decision levels leads to unique challenges and opportunities, such as aligning measurements in time, managing missing observations, and accounting for the high correlation between and within the data streams. In addition, each piece of information has separate uncertainty, which must be properly combined and propagated to the final result. Also of interest is the value each data modality adds and the minimum and correct set of data necessary to answer a question with a defined amount of precision. No single data fusion algorithm can address all the potential issues, but many problems have been studied in significant depth and will be discussed in this session.
|