UC Irvine

This dissertation proposes a set of coherent cognition-based paradigms to allow greater sensitivity and adaptability to the emerging technologies and behavioral policies. These paradigms are derived from a cognition-based framework that explicates information source, medium, sensation, perception, and learning. The feasibility of the framework is demonstrated through an analytical example of multi-stakeholder decision processes and human-machine systems where the two types of entities can be incorporated into the same modeling scheme. Using the framework as guidance also reduces the challenges from information intractability and data redundancy of agent-based modeling practice.

The first paradigm follows the strict definition of information in Information Theory and models it as the change of uncertainty, which is applied in quantifying traveler information for the evaluation of dynamic message boards that present various contents at candidate locations in Los Angeles traffic networks.

The second paradigm is developed for a utility-based decision model under risk around the proposed concept, Elastic Surprise. This concept makes feasible the differentiation between probability misperception and perceived uncertainty. It is shown that conventional methods of decisions under risk such as Expected Utility Theory and Cumulative Prospect Theory are special cases. In addition, a specific form of Elastic Surprise under particular assumption on human's cognition leads to Shannon's information entropy and, hence, connects with the first Paradigm. The method is tested in conjunction with the Cumulative Prospect Theory on travel time equivalency under risk in a survey study. The results show improvement in data fitting and output interpretability.

Finally, guided by the framework, the paradigms are tested on a case study of multi-class multi-criteria dynamic traffic assignment where heterogeneous travelers' risk preference on travel time is explicitly modeled. The algorithm approaches the user equilibrium through a stochastic quasi-gradient projection-based algorithm that shows the improvement in computational efficiency and cognitive implication of the agents’ decision rules. I also discuss the potential strategies and policies implication for system improvement.