A University of Maryland researcher is part of a Multidisciplinary University Research Initiative (MURI) that seeks to overcome one of artificial intelligence's biggest limitations: combining the adaptability of machine learning with the reliability of logical reasoning. The work could lead to more dependable autonomous systems and help AI tools make new scientific discoveries by drawing more reliable conclusions from complex data.
The five-year, $7.5 million project includes approximately $1.4 million for UMD, contingent on continued funding, to support research led by Soheil Feizi, an associate professor of computer science with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS).
The project, SOLIR: Stochastic Ontology for Learning: Information Theory Meets Rule-Based Reasoning, is funded by the U.S. Department of Defense and led by Rutgers University. Salim El Rouayheb, an associate professor of electrical and computer engineering at Rutgers, serves as the principal investigator. In addition to Rutgers and UMD, the multidisciplinary team includes researchers from the University of Illinois Chicago, the Massachusetts Institute of Technology and Drexel University.
“Our goal is to build a mathematical foundation for AI systems that can both learn from data and reason with structured knowledge,” said Feizi, who is also a core faculty member in the UMD Center for Machine Learning. “By combining information theory with structured reasoning, we hope to create AI systems that are more reliable, efficient and transparent, especially in settings where decisions must be made under uncertainty. These advances could improve everything from autonomous systems to large-scale knowledge networks while helping ensure that AI can explain not only what it predicts, but how it arrives at those conclusions.”
MURI awards are among the Department of Defense’s most competitive basic research programs, supporting large, multi-institutional teams pursuing high-risk, high-impact research with the potential to strengthen future national defense capabilities.
The SOLIR project brings together two traditionally separate approaches to artificial intelligence. Modern AI systems excel at learning patterns from vast amounts of data but can produce unreliable or difficult-to-explain results. Rule-based systems, by contrast, rely on explicit logical reasoning but often struggle with uncertainty and incomplete information. The researchers aim to unite the strengths of both approaches by developing mathematical frameworks that enable AI systems to reason under uncertainty while remaining grounded in structured knowledge.
At Maryland, Feizi will co-lead the project's first research thrust, focused on developing the information-theoretic foundations for structured knowledge, or ontologies, in AI systems. His research will examine how AI can identify and represent the most relevant concepts, relationships and rules for a given task, enabling systems to reason more efficiently while preserving the information needed for accurate and reliable decisions.
A key focus will be understanding how this structured knowledge can be represented, updated and compressed without losing important information needed for reasoning. The research will also examine distributed settings, where multiple AI systems must share and act on evolving knowledge under constraints such as limited bandwidth, energy and computing resources.
Over the course of the project, the research team will develop open-source software to demonstrate the new algorithms and mathematical theories. The software will be evaluated using representative applications, including autonomous systems and emergency response scenarios, while providing a foundation for collaboration across the project's research thrusts.
Throughout the five-year project, Feizi will collaborate closely with researchers developing complementary approaches to stochastic ontology models and decentralized learning algorithms. Together, the team aims to establish a foundation for AI systems that can learn from data, share knowledge efficiently and reason more reliably—even when information is incomplete or uncertain.
—Story by UMIACS communications group