CAREER: Towards a Sensorless Internet of Things through Sensor Data Management Abstraction

The Internet of Things (IoT) refers to the billions of sensors and connected devices embedded in buildings, factories, hospitals, and cities that continuously collect data about the world around us. Despite the enormous potential of this data to improve how we manage energy, monitor public health, and run our industries, most organizations struggle to make use of it. The reason is that turning raw sensor readings into useful answers requires deep technical expertise. This includes knowing which sensors to consult, which AI models to apply, and how to combine everything correctly. This project addresses this challenge by developing a new approach that allows anyone to simply ask what they want to know (e.g., "which rooms were underused last month and how much energy did they consume?") while the computer system intelligently figures out how to find the answer. By hiding the complexity of sensors, this work has the potential to make the benefits of IoT technology to a much wider range of Americans, including educators, building managers, public health officials, and factory operators, through data management and artificial intelligence. The project also trains the next generation of students through hands-on IoT courses at the university level, high school internships, and a summer camp for students in the Baltimore area. This project advances the foundational science of IoT data management by introducing semantic abstraction as a first-class concept in query processing. Rather than requiring users to specify the sensors, models, and data pipelines needed to answer a question, the system treats high-level concepts (such as occupancy, air quality, or energy consumption) as directly queryable entities. The research is organized around three technical thrusts. The first defines a formal query model and optimization framework that reasons about multiple ways to compute a desired concept, selecting among them based on accuracy, latency, and resource cost. The second develops self-driving abstraction planning techniques that learn, using methods from machine learning and adaptive scheduling, when to compute abstractions eagerly at data ingestion time versus lazily at query time, and how to execute them efficiently across heterogeneous storage systems including time-series, relational, graph, and document databases. The third thrust extends the framework to federated settings where data are spread across multiple organizations with different privacy policies, enabling incremental, confidence-annotated answers under partial data availability and access constraints. All techniques will be validated in real-world deployments including a smart campus environment at the University of Maryland, Baltimore County, a smart manufacturing testbed, and a smart home laboratory. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2542782 | Program: 01002930DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Roberto Yus | Institution: University of Maryland Baltimore County, BALTIMORE, MD | Award Amount: $419,429 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2542782 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2542782.html