Testing and Verification of Deep Neural Network

In AI safety, compliance ensures that a model adheres to operational specifications at runtime to avoid adverse events for the end user. This proposal looks at increasing the suite of compliance testing and verification tools available to ML practitioners by implementing two more tools, CGDTest and Goose. The overall goal is to develop both proposed tools to be flexible in real-world applications, and the tools can be evaluated by submitting to an annual competition for neural networks verification (VNN-COMP). VNNCOMP is a highly recognized competition often used as a benchmark for testing and verification algorithms in the community. Real-world applications for such tools include their usage in ML model validation pipelines to minimize operational and reputational risk when deploying ML models in production in large financial institutions.

Faculty Supervisor:

Vijay Ganesh

Student:

Partner:

Royal Bank of Canada (Borealis)

Discipline:

Computer science

Sector:

Information and Communications Technology; Technology; Finance and Insurance

University:

University of Waterloo

Program: