Jaechul (Harry) Roh
I'm an upcoming final year undergraduate student in computer engineering at Hong Kong University of Science and Technology (HKUST) in the School of Engineering. I began my journey on studying the field of adversarial attack by choosing such topic as my machine learning course presentation project. I was fascinated by how a simple and swift algorithm such as Fast Gradient Sign Method (FGSM) can both fool and strengthen any deep neural network.
2022/04/15: Paper "Impact of Adversarial Training on the Robustness of Deep Neural Networks" (paper ID: MSCS-704), which I have been working on during the 2021/22 Winter holiday, has been recently accepted by the 2022 International Conference on Modeling, Simulation and Computing Science (MSCS 2022). It will be published by IEEE CS (Computer Society) CPS and submitted to EI database for indexing.
2022/06/01: Completed independent work research on the topic of “MSDT: Masked Language Model Scoring Defense in Text Domain” under Professor, Minhao Cheng, where we proposed a novel improved textual backdoor defense algorithm that outperforms the state-of-the-art ONION method in specific datasets.
2022/05/23 - Present: Currently participating in the IEEE International Conference on Universal Village. Experimenting the robustness of federated learning in smart home face recognition system in respect to MIT Universal Village (UV) concept under the supervision of Dr, Yajun Fang (CSAIL, MIT).