Johnstown-based Concurrent Technologies Corp. (CTC), an independent, nonprofit, applied scientific research and development professional services organization, has partnered with the Robotics Research Center at the United States Military Academy, West Point, on a research project.
The project will design and engineer a machine learning model-stealing pipeline that will become part of an AI Security Assessment Tool that can automatically test and evaluate tools and metrics. The pipeline will address critical security vulnerabilities in artificial intelligence systems. CTC also is conducting scientific research using the pipeline.
“This partnership with West Point’s Robotics Research Center represents a significant step forward in securing AI systems against emerging threats,” Dave Saranchak, CTC executive director, said. “Our team is proud to contribute our expertise in developing tools that will help protect critical machine learning models from adversarial attacks and unauthorized access. This work is essential to maintaining the integrity and security of AI systems that our nation depends on.”
Potential applications for the pipeline include identifying specific model components or features most vulnerable to exploitation, identifying which architectures and training techniques are more susceptible to privacy leaks, and evaluating privacy-preserving techniques that make it harder for attackers to infer hidden characteristics of training datasets.