Abstract:
The existence of evasion attacks during the test phase of machine learning algorithms repre-
sents a signi cant challenge to their deployment and understanding. These attacks are carried
out by adding imperceptible perturbations to the inputs to generate adversarial examples. As
of now designing good robust classi ers in real life seems very di cult. But so far most of the
studies depict the relationship between computational power of adversary and robustness of
the classi er. In this report, we have used some of the cryptographic schemes to create robust
classi ers and show the dependency of robustness with adversarial budget.
Description:
Dissertation under the supervision of Dr. Dr. Debrup Chakraborty,Associate Professor,Cryptology and Security Research Unit