Abstract not found

For decades, researchers have debated the pros and cons of adaptive user interfaces with enthusiastic AI practitioners often confronting skeptical HCI experts (Shneiderman & Maes, 1997). This paper summarizes the SUPPLE project’s

For decades, researchers have debated the pros and cons of adaptive user interfaces with enthusiastic AI practitioners often confronting skeptical HCI experts (Shneiderman & Maes, 1997). This paper summarizes the SUPPLE project’s

I am broadly interested in human computer interaction, machine learning and artificial intelligence. My dissertation demonstrates how to automatically generate personalized adaptive user interfaces. My central thesis is that personalized user interfaces, which are adapted to a person’s devices, tasks, preferences and abilities, can improve user satisfaction and performance. Further, I demonstrate that automatic generation of personalized user interfaces is computationally feasible. In my dissertation work, I developed three systems to enable a broad range of personalized adaptive interfaces: SUPPLE, which uses decision-theoretic optimization to automatically generate user interfaces adapted to a person’s device and usage pattern; ARNAULD, which allows optimization-based systems to be adapted to users ’ preferences; and SUPPLE++, a system that first performs a one-time assessment of a person’s motor abilities and then automatically generates user interfaces adapted to that user’s abilities. The results of my laboratory experiments show that these automatically-generated, ability-based user interfaces significantly improve speed, accuracy and satisfaction of users with motor impairments compared to manufacturers ’ defaults. I also provide the first characterization of the design space of adaptive graphical user interfaces, and demonstrate how such interfaces can significantly improve the quality and efficiency of daily interactions for typical users. SUPPLE: A Platform For Automatically Generating Personalized User Interfaces My SUPPLE system [9] uses constrained decision-theoretic optimization to automatically generate user interfaces. As input, SUPPLE takes a functional specification of the interface, which describes the types of information that need to be communicated between the application and the user, the device-specific constraints, such as screen size and a list of available interactors, a typical usage trace, and a cost function. The cost function can correspond to any measure of interface quality, such as expected ease of use or conformance to user’s preferences. SUPPLE’s optimization algorithm finds the user interface that minimizes the cost function while also satisfying all device constraints. Despite the huge space of possible designs (exponential in the number of