Main Download Video Demos
PRODeep is a platform for robustness verification of deep neural networks (DNNs). It incorporates constraint-based, abstraction-based, and optimisation-based robustness verification algorithms. It has a modular architecture, enabling easy comparison of different algorithms.
Significantly, PRODeep provides a user-friendly GUI, visualising both inputs and outputs and providing an intuitive way to analyse the robustness properties. It is easy to get started with, so you can easily design some experiments to evaluate the robustness properties of your DNNs.
The screencast on YouTube will show the workflow of our tool.
Please feel free to contact us for any further information on PRODeep you might need.
Analyzing Deep Neural Networks with Symbolic Propagation: Towards Higher Precision and Faster Verification
, Li, J.; Liu, J.; Yang, P.; Chen, L.; Huang, X. and Zhang, L. In SAS, pages 296-319, LNCS 11822, 2019.
Improving Neural Network Verification through Spurious Region Guided Refinement.
Yang, P., Li, R., Li, J., Huang, C. C., Wang, J., Sun, J., Xue, B., Zhang, L. (2020). arXiv preprint arXiv:2010.07722.
Reluplex: An efficient SMT solver for verifying deep neural networks.
Katz, G., Barrett, C., Dill, D. L., Julian, K., & Kochenderfer, M. J. (2017, July). In International Conference on Computer Aided Verification (pp. 97-117). Springer, Cham.