Researchers have developed a new testing framework called TESTNAV to identify vulnerabilities in deep learning models. The framework uses a Pareto-guided approach to efficiently explore the space of possible input perturbations, prioritizing those that cause severe yet realistic failures. This is achieved by formulating robustness testing as a bi-objective optimization problem, where both performance degradation and input fidelity are considered. TESTNAV has been shown to recover Pareto fronts up to 2.15x faster than existing search-based methods, using a fraction of the perturbation space defined by four dimensions with six levels each. --- Why it matters: This matters because deep learning models remain vulnerable to real-world input perturbations, and compositional testing is essential for revealing interaction effects between multiple corruptions. TESTNAV's efficient exploration of perturbation spaces can help researchers identify and mitigate these vulnerabilities more effectively. Source: https://arxiv.org/abs/2608.19882