Modeling of human information processing in hard-soft fusion (Poster)

With the advent of human-in-the-loop systems, it is necessary to design machines that are optimal for human use. For this purpose, it is important to understand and build models of human behavior. In this paper, human information processing is modeled as a communication channel and information-theoretic metrics are used to quantify amount of information processed by humans in a two-choice inference task. Single-shot inference and sequential inference frameworks are considered and the expected information transfer is derived for human decision makers (HDMs). The results are also extended to derive upper bounds on expected amount of information transfer for the case of a network of HDMs. Ordering principles based on these metrics are presented. Such representations of soft sensors' data guide the design principles of hard sensors and the hard-soft fusion algorithms.