Falsification-Driven Robust Training

Using adversarial falsification to expand certified L2 robustness radii.

Built a pipeline coupling the CARLA simulator with VerifAI to synthesise adversarial examples, then used them to train models that are provably more robust.
Computed certified L2 robustness radii with CROWN via auto_LiRPA, chosen over the tighter Alpha- and Beta-CROWN variants whose VRAM and latency demands made verification infeasible at this scale on a 6 GB GPU.
Falsification-driven active learning expanded certified L2 robustness radii over three retraining cycles, showing markedly better sample efficiency than random sampling.