Machine learning models have demonstrated remarkable success across diverse
domains but remain vulnerable to adversarial attacks. Empirical defence
mechanisms often fall short, as new attacks constantly emerge, rendering
existing defences obsolete. A paradigm shift from empirical defen