Federal Project Grant Award Summary

Worcester Polytechnic Institute received a $369,231 CAREER award from the National Science Foundation's Division of Computer and Network Systems (CFDA 47.070: Computer and Information Science and Engineering) beginning May 1, 2026, with completion targeted for April 30, 2031. This research project addresses critical security vulnerabilities in privacy-preserving neural networks deployed at the edge by developing a comprehensive vulnerability assessment methodology. The primary deliverables include the development and evaluation of physical-security-first approaches that combat three distinct threat vectors: backdoor insertion during model training, information leakage during inference through non-invasive and semi-invasive attacks, and fault injection attacks combined with machine learning techniques to infer secret model parameters and architecture.

Beyond technical research outputs, the project delivers significant broader impact products including a public vulnerability issue database and vulnerability-discovery competitions designed to train next-generation engineers in secure neural network system development and evaluation. The initiative directly advances national security and trust in sensitive applications by reducing risks associated with model manipulation, intellectual property theft, and user data privacy violations. These educational and community dissemination components position the research to enhance workforce development while establishing measurable advances in the security posture of privacy-preserving neural network deployments across critical real-world applications including healthcare wearables, smart homes, and autonomous vehicles.