SBIR/STTR Award attributes
In Phase 1, we researched on the current state of the art on how to build a robust and safe AI/ML applications for CNN,RNN and RL based applications. We studied the important features and critical metrics of robustness and safety during ML model training, testing, model deployment and production and tested our initial proof of concept. In Phase 2, we will continue to develop the technical framework of CORSI, build a functional prototype of CORSI components, identify areas of optimization & performance bottlenecks, create a production-ready robustness and safety pipeline, identify areas of weakness & enhancements in the opensource tools and develop CORSI prototype. By utilizing our Risk Framework and Assumptions & Safety Violation Framework, CORSI will generate a Certificate of Robustness and Safety (CORS) as the final outcome for each application. In Phase 2, our goal is to build a functional, easy-to-understand, scalable and responsive CORSI prototype that can handle medium to complex use-cases to validate the robustness and safety of AI applications. We will give specific emphasis on building a simplified, scalable and high-performance framework with the following technical objectives: Develop and enhance the system architecture and technical framework of CORSI components with specific emphasis on ease of implementation, scalability, performance measurement; Identify major challenges in phase 2 implementation of a scalable CORSI to handle medium to complex use-cases and continually monitor emerging technologies in rapidly moving AI/ML landscape; Identify general-purpose unclassified use-cases with varying levels of complexity; Develop a plan for training and test data collection, robustness and safety metrics and performance metrics of CORSI; Develop and execute system/module test plan for CORSI components and robustness/validation test plan.
