An AI Assurance research program requires a testbed, consisting of high performance computing, a software factory, and test suite. The AI Assurance Testbed will be a collaborative activity and joint resource of the Commonwealth Cyber Initiative (CCI). Principal development of the core hardware and software and expertise in standing up the equipment will be provided by the CCI Hub. CCI Nodes will explore research issues related to specific applications. Hub and Node researchers will access the testbed through remote user accounts.
The testbed will leverage an NVIDIA GBX-2, an enterprise-grade research computer designed for AI, machine learning, and deep learning applications, and 200 TB of storage. Staffed by experts, the AI assurance testbed will have pre-trained open source models for easy test case execution, covering a range of AI algorithm types from support vector machines to deep neural nets, and a test suite of baseline options for various functional tests.
As use cases evolve, this testbed will be augmented with additional computing hardware, storage, data, algorithms, and test cases.
The AI Assurance Testbed and Software Factory built on the DGX-2 will:
- Provide a high-quality and high-efficiency test and experiment facility for academic research by academic faculty, postdocs and PhD/MS graduate students in both the CCI Hub and Nodes.
- Provide a test capability for red and blue-teaming AI algorithms for (a) students and academic researchers (b) industry collaborators and (c) government sponsors working in AI research and development.
- Provide a vulnerability testing and validation service for government T&E organizations of COTS and GOTS algorithms.
- Serve as a resource for other research areas that will likely be dependent on AI Assurance. For example, AI considerations in 5G wireless technologies or cyber-physical systems. The testbed will serve as an initial software test facility for these application areas.