A former schoolhouse more than 90 years old, the structure was purchased in 1991 by Associate Professor Michael Mozer of the University of Colorado at Boulder's computer science department. It was then renovated and retrofitted with high-technology hardware. Using data gleaned by sensors installed by Mozer and his students, the computer system essentially "programs itself" by observing his lifestyle and habits over time, eventually learning to anticipate and accommodate his needs.
Mozer and more than a dozen graduate and undergraduate students have installed 75 sensors and nearly five miles of conductor in the home, as well as actuators to control lighting, ventilation and air and water heating. The sensors continually monitor temperature, ambient light, sound and motion in each room, the opening of doors and windows, outdoor environmental conditions, boiler temperature and hot water usage.
Many homes can be programmed to perform tasks like watering lawns or turning on televisions, but programming a home is a complex and difficult task that few homeowners are interested in doing, he said.
"The twist is that this house programs itself by observing inhabitants as they live their lives," he said. "The system is based on neural networks, which are learning devices inspired by the working of the human brain."
 |
| Click here to download high resolution .jpg |
The human brain relies on billions of neurons constantly communicating with each other as they acquire knowledge and form memories. In Mozer's house, artificial neural networks consisting of hundreds of simple, neuron-like processing units interact to predict and control the environment.
The system predicts Mozer's behavior and movements, including which rooms will be occupied at what times, when he will leave the house and return, and when hot water will be needed in the boiler.
"The system infers rules of operation and adapts to the lifestyle of the inhabitant," maximizing comfort by setting appropriate temperatures and light levels while minimizing energy consumption," he said.
In Mozer's house, anticipating and carrying out the wishes of the inhabitant and conserving energy sometimes conflict. So Mozer and his colleagues at CU's Institute of Cognitive Science devised mathematical techniques for translating discomfort to a cost in dollars that can be weighed against energy costs.
One technique, based on an economic analysis, depends on the loss in productivity that occurs when the system ignores the inhabitants' desires. Another technique adjusts the relative importance of the inhabitants' desires based on how much they are willing to pay for gas and electricity.
Even if the inhabitants do not have a particularly regular schedule, there are statistical regularities in their behavior that the system can exploit. For example, if Mozer is not home by 3 a.m., he likely will not be home by 4 a.m. and therefore the house does not need to be warmed up.
Mozer demonstrated the bathroom light, which turned on to a low intensity as he entered. "The system picks the lowest level of the light or heat it thinks it can get away with in order to conserve energy, and I need to complain if I am not satisfied with its decision," he said.
To express his discomfort, he hit a wall switch, causing the system to brighten the light and to "punish itself" so that the next time he enters the room, a higher intensity will be selected.
"The house has been the source of a dozen student research projects and two masters theses," he said. "This is a good testing ground for undergraduates who have never actually built something in the real world. And we are in a domain that I believe may have great practical potential."
Much of Mozer's neural network research has been funded by the National Science Foundation. In 1990, Mozer received a prestigious Presidential Young Investigator Award from NSF worth $100,000 annually for five years. The NSF also funded some of Mozer's undergraduates through its Research Experience for Undergraduates program.
- 30 -