Adaptive Interfaces in Complex Supervisory Tasks

Abstract

We have proposed a new method for adaptation in user interfaces. Under the assumption of users stationary behavior for an interface we have shown that combination of GA (Genetic Algorithm) and probabilistic modeling of user may refine the interface to the point of personalization. Non-parametric statistics has been employed in order to evaluate feasibility of our ranking approach. Method proposed was flexible and easy to use in order to be applied to any problem domain. Our automated user was developed under the assumption that a- limited cognitive and motor abilities and b- stationary probabilistic, same as our assumptions about the user. These assumptions were considered in the past as in many different papers. Automated user was employed to show the convergence of the algorithm in large amount of interface parameters. We have also shown that using performance metrics for adaptation leads to adaptation in physiological space of the human users. Difference between automated user and human user in the interface model sense was the size of the search space. We have designed a smaller search space for user due to user burden. Results form the automated user and the real user was consistent and shows that algorithm is capable of handling both large search space and the real problem domain.