Created and published an adversarial ML toolbox 'deepillusion'
Combating adversarial perturbations by exploiting sparsity of natural data.
Teaching a network to play Turkish card game “Batak” by using Reinforcement Learning methods. Asynchronous Advantage Actor Critic (A3C) algorithm is implemented for learning. Game environment for agents is written from scratch. LSTM based network model is used to capture temporal information.
Extracting circuit-level fingerprints to distinguish between wireless devices sending exactly same message