Sensor networks for discovery

Summary form only given. Advances in sensors, signal processing, and communications have opened the possibility of designing systems composed of dense spatial networks of low-cost sensor/processor elements working in concert to measure and analyze complex spatio-temporal fields and patterns. This possibility presents new challenges to our understanding of the capabilities, limitations, and design tradeoffs of networked collections of sensing and processing elements. We are investigating a new algorithmic framework that allows us to optimize jointly all aspects of sensor network operation - data collection, model selection, data processing, and communication - to provide systems that are capable of self-configuring in response to data in order to adapt and optimize their capabilities to understand the physical environment they are sensing. A crucial aspect of our proposed framework is that it places as few constraints as possible on the sensor network prior to collecting data. As measurements are made, the sensor network configures its structure, models, processing, and communication to the environment. At present, the framework is only a conceptual one, but we anticipate that the basic principles of our framework, which build on recent advances in coding and statistical learning theories, may have important implications for future sensor networks.