Finding Adversarially Robust Graph Lottery Tickets
The goal of this project is to find a adversarially robust graph sparsification technique to adress the problem that the performance of GLTs collapses against structure perturbation poisoning attacks. By iteratively applying ARGS, we found ARGLTs that are highly sparse yet achieve competitive performance under different structure poisoning attacks. We prune the perturbed adjacency matrix and the GNN weights by optimizing a novel loss function. Our evaluation showed the superiority of our method over UGS at both high and low-sparsity regimes.
January 2023 - Present
Risk-Aware Cost-Effective Design Methodology of Logic Locking
The goal of this project is to propose a risk-aware IC locking methodology which enables systematic and efficient exploration of a large range of design alternatives via rigorous quantification of the achievable protec-tion levels, residual risk, and their trade-offs with area overhead. By combining a set of locking primitives, it is possible to generate new constructions that outperform existing methods, used in isolation.
August 2022 - Present