Consider a demand response provider that wants to learn a personalized policy for each user, but the platform faces the risk of a user abandoning the platform if she is dissatisfied with the actions of the platform. For example, the platform will want to personalize the thermostat control for the user, but faces the risk that the user unsubscribes forever if they are mistreated. We propose a general thresholded learning model for scenarios like this, and discuss the structure of optimal policies.

Abstract: Scientific challenges: How can multiple cooperative cyber-physical systems communicate and coordinate to accomplish complex high-level tasks within unknown, dynamic and adversarial environments?

Abstract: In order to improve the current capabilities of automotive active safety control systems (ASCS) one needs to take into account the interactions between driver/vehicle/ASCS/environment. To achieve this goal, this research will infer longterm and short-term driver behavior via the use of Bayesian networks and neuromorphic algorithms to estimate the driver’s skills and current state of attention from eye movement data, together with dynamic motion cues obtained from steering and pedal inputs.

Abstract: One defining feature of cyberphysical systems is the fact that human users are closely intertwined with the physical system. Yet both the system and users themselves are often do not explicitly know how users would behave. A natural question arises: How do we design cyberphysical systems that effectively learn about their users, and optimize system behavior accordingly? This poster presents the idea of active regression as a vehicle to learn about users.