Federal Project Grant Summary

The National Science Foundation (NSF) Division of Computer and Network Systems awarded a $344,911 CAREER grant to Rochester Institute of Technology (RIT) on April 1, 2026, for a five-year project (through March 31, 2031) addressing adversarial threats to machine learning (ML) systems. Under the Computer and Information Science and Engineering (CFDA 47.070) program, this award funds research to develop unified interpretability and attribution frameworks that enhance the security and resilience of ML models deployed in critical sectors including healthcare, finance, and national security. The primary deliverables include scalable attribution techniques combining gradient-based influence methods with symbolic surrogate models such as decision trees; fast localized model editing strategies to remove harmful behavior without complete retraining; and automated pipelines for training data auditing, provenance tracking, and security-aware valuation to detect data poisoning.

Beyond core research objectives, the project delivers substantial educational and practical impact through hands-on laboratory curricula, workforce development training in cybersecurity-aware machine learning, and tools enabling practitioners to interpret and defend complex ML systems. The four-thrust technical approach bridges classical, interpretable frameworks with modern complex systems, allowing defenders to trace failure origins, validate repairs semantically, and build defensible ML architectures. This work directly supports NSF's mission to advance investigator-initiated research in computing and information science while strengthening the accountability and safety of ML systems in high-stakes operational environments.