The goal of this project is to develop a program that can synthesize an abstraction of any classification neural network using counterexample-guided inductive synthesis (CEGIS) and the Marabou neural network verification tool (neuralnetworkverification.github.io).
As of May 2024, the project is able to synthesize unrefined piecewise abstractions of binary classification neural networks with some inefficiencies and limited testing. Please see todo.md for next steps.
To be determined once development is complete.
Please read the CONTRIBUTING.md file in the docs directory for steps on how to contribute to the development of this project.
Please read the info.md file in the docs directory to understand how this project directory is structured and what its contents are.