Location Prediction for Sustainability

My research interest lies in using applied Machine Learning, especially Location Prediction, to find solutions for the ever growing resource consumption problem of our society. By predicting a user's movements through open and closed spaces it will be possible to build systems and applications that support Sustainability. Possible applications can be but are not limited to: automatic control of appliances, lights, or temperature regulation, but also pervasive systems that aim at changing a user's behavior.

My research uses lessons learned in Machine Learning and combines them with Human-Computer-Interaction (HCI) methods to not only explore, design, and develop sustainabile solutions that have an impact on current consumption, but also to make sure that the developed systems are well accepted by users.

To satisfy the latter goal I'm also intersted in how Machine Learning and ambient sensors can be used to detect a user's current comfort level and predict the impact of changing environmental variables (light, temperature, etc.) on the comfort of users. To achieve this it is important to understand impact factors on a person's comfort level and find ways to not only measure them, but also design systems that learn these highly individual factors.


Software Framework for SINAIS

framework During an internship I did at the Madeira Interactive Technologies Institute in Spring/Summer 2010 I developed a software framework that allows coordination of distributed sensors. The framework features a layered structure with an XML based communication protocol between these layers. Through adoption of the Blackboard pattern the framework not only allows processing of sensor events, but more importantly comparison of different processing algorithms against each other. The framework is currently used to gather power consumption data from a novel low-cost power meter developed by researchers of the CMU|Portugal funded SINAIS project.


Automatic Thermostat Control and Behavior Modification

framework This project is an ongoing effort resulting from my Master thesis in 2009. Goal is to explore the use of outdoor location prediction to automatically control the thermostat in apartments and investigate the impact on power consumtion resulting from heating and cooling. By offering the user control over the system and consumption feedback through a mobile phone application I'm also interested in how mobile phones can influence a user's behavior. Techniques employed for this behavior modification are just-in-time recommendations, goal-setting, and behavior reflection through daily, weekly, and monthly graphs.

Special thanks to Brian Ziebart, a CMU Ph.D. student, who's work in predictive models made the automatic part of the project possible.

[pdf] Koehler, C., Dey, A., Mankoff, J. and Oakley I. "Motivate Environmentally Sustainable Thermostat-Use through Goal-Setting, Just-In-Time Recommendations, and Behavior Reflection", submitted to NIMD'10, Mobile HCI 2010, Lisbon, Portugal