Defense against Adversarial Attacks on Deep Neural Networks

I did my undergraduate thesis in very intresting topic in the supervision of Dr. N. B. Puhan. We picked up the challenging topic of proposing a novel adversarial defense against the adversarial attacks on Deep Neural Nets in the field of Computer Vision.

As part of this project we explored different techniques ranging from usage of Stirmark software, Collusion based defense, Denoising methods, etc. And finally came up with some novel methods Like NOMARO, and Local RPCA defense.

Below PDF contains is the final thesis report of my project and contains fundamental details on every defense approach taken. 

Although post this we explored more on different DNN Models and made NOMARO, and RPCA defense more efficient and finally published NOMARO defense in IEEE Sensors Letters (2021) and have presented our work on Local RPCA defense in IEEE ICORT 2023 conference. 

More details on these defenses can be found in subpages - NOMARO defense, Local RPCA defense