SBIR/STTR Award attributes
By leveraging open-source tools and frameworks, we propose to build CORSI - a suite of AI verification & validation tools of different types of neural networks operating under different circumstances. By utilizing our Risk Framework, Assumption and Safety Violation framework, CORSI will generate a Certificate of Robustness (COR) as the final outcome. Our tools will have 3 components DNN Toolkit, Bayesian Uncertainty Modeling and Probabilistic Verifier covering CNN. CORSI will identify areas of concern, exceptions, red flags & vulnerable areas and provide remedial action. Once the AI system goes into production, CORSI's Intelligent Run-time Agent (IRA) will provide 24 x7 monitoring to evaluate the production system and provide protective measures and insights.
