This is a project which is currently making use of HPC facilities at Newcastle University. It is active.
For further information about this project, please contact:
This project will develop and evaluate approaches for the robustness analysis of deep neural networks. The problem is mainly about the accurate and efficient measurement of robustness metrics including the Lipschitz constant and worst-case classification margins. The methods are mainly formulated as Semidefinite Programming (SDP) or iterative first-order methods. HPC resources are used to accelerate these complex computations, enable benchmarking of the developed approaches against baselines, and scale robustness analysis beyond local compute limits.
The project will use GPU-accelerated deep learning frameworks such as PyTorch, alongside convex optimization tools including CVXPY, the MOSEK solver, and specialized neural network verifiers including auto_LiRPA. HPC GPUs will be essential for first-order methods, while high-memory compute nodes will support the MOSEK solver in processing memory-intensive LMI formulations.