Abstract
To study the Adversarial attacks and Defence techniques for Machine Learning models. I conducted an exhaustive literature survey on state-of-the-art adversarial sample generation techniques and defense methods for DNNs. I successfully developed Non-targeted adversarial attacks and formulated reactive and proactive defence techniques for improving the robustness of visual question answer model TGIF-QA. Future work involves developing an effective defense method with high success rate on most attacks.
Ganga Meghanath
Data & Applied Scientist
My research interests include Reinforcement Learning, Deep Learning, Game Theory, Vision & Robotics.