NMSU wins $600,000 DARPA grant to develop AI that can show its reasoning
The two-year SPARTA project will combine machine learning with automated reasoning. Verifiable explanations and reusable methods are research goals, not demonstrated results.
Loading page…
The two-year SPARTA project will combine machine learning with automated reasoning. Verifiable explanations and reusable methods are research goals, not demonstrated results.
Listen to this story
The NMSU-led SPARTA effort will investigate ways to pair machine-learning pattern recognition with logic-based reasoning, so AI outputs can be checked against formal proofs without making the computation impractical. The two-year award is part of DARPA’s CLARA program, which funds foundational work toward reusable methods for complex, high-stakes systems. The work is research, not a finished product; its value will depend on whether the team can make verification both scalable and applicable across different problems.
NMSU says it is one of 10 research teams nationwide receiving awards through the DARPA program.
Computer science professors Son Tran and Stylianos Vasileiou lead NMSU’s effort.
The collaboration includes a group led by Chitta Baral at Arizona State University and another led by Yuliana Lierler and Jorge Fandinno at the University of Nebraska at Omaha.
New Mexico State University researchers have received a two-year, $600,000 DARPA grant to develop AI whose answers come with reasoning people can verify. Announced November 20, the award supports an effort to combine machine learning’s ability to recognize patterns with automated reasoning, which uses logic to reach conclusions.
Computer science professors Son Tran and Stylianos Vasileiou lead the NMSU effort. Tran is the department head; Vasileiou is an assistant professor. The university says their team is one of 10 nationwide receiving grants through the DARPA research program.
Their proposal is called SPARTA: Splitting-based Polynomially-bounded Automated Reasoning with Tensorized Adaptation. It sits within CLARA, short for Compositional Learning-And-Reasoning for Automated Complex Systems Engineering. The Defense Advanced Research Projects Agency program funds fundamental research into tightly integrating learning and reasoning components to create machine intelligence with stronger assurances of reliability.
Tran says the team will develop efficient, scalable ways to combine machine-learning approaches with logic programs.
Integrating these branches of artificial intelligence will provide the speed and flexibility of machine learning with verifiability based on artificial reasoning proofs that have strong logical explainability and computational tractability
Son Tran, NMSU computer science professor and department head
That last requirement—computational tractability—means keeping the reasoning manageable to compute. The team’s stated goal is to develop methods that are both broadly applicable and capable of supporting strong assurances about a system’s behavior.
NMSU is working with two other research groups: one led by Chitta Baral at Arizona State University, and another led by Yuliana Lierler and Jorge Fandinno at the University of Nebraska at Omaha. Their collaboration forms part of the SPARTA research effort described in the university’s announcement.
The larger CLARA ambition is to reduce the custom engineering needed to make AI suitable for critical work. NMSU describes a program seeking a mathematically grounded foundation that can adapt to many complex problems, rather than requiring developers to rebuild a system for each application.
The intended destination is a reusable AI engine for high-stakes industries. For NMSU and its collaborators, the funded task is foundational research: developing the methods that could make that reuse possible.
Loading discussion...
Join the conversation
Explain where reuse makes sense and where a purpose-built system would be better.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.