Enhancing interpretability of deep neural networks through topological analysis using reeb graphs

The University of California, Santa Barbara (UCSB) Office of Research received a $222,448 Cooperative Agreement award from the Research Development and Engineering Command (RDECOM), a defense agency component, under the Basic Scientific Research program (CFDA 12.431). The award was obligated on August 1, 2025, with a performance period extending through July 31, 2028. The research initiative focuses on "Enhancing Interpretability of Deep Neural Networks Through Topological Analysis Using Reeb Graphs," representing fundamental scientific investigation in the computing and mathematical sciences domain to support long-term national security needs and advance the Department of the Army's technological capabilities.

Under this three-year cooperative agreement, UCSB will conduct research addressing the critical challenge of artificial intelligence (AI) explainability by developing novel analytical methods using topological approaches. The project delivers scientific and technical advancement in deep neural network interpretability, a high-priority research area essential for trustworthy AI systems in defense applications. As a major research institution with demonstrated capabilities in machine learning and advanced computational research, UCSB will perform this fundamental research at its Santa Barbara, California facility, contributing to the Army's strategic objectives in emerging technology domains while advancing broader scientific knowledge in AI and topology.