Sanghyun Hong 홍상현 메릴랜드 대학교

Ph.D Candidate in Computer Science
at University of Maryland, College Park, advised by Prof. Tudor Dumitraș.

Research Interests
My research objective is to solve the security and privacy problems of machine learning (ML) systems. For this, I am particularly interested in characterizing the vulnerable interactions between ML algorithms and their surrounding environments such as hardware/systems where ML algorithms are deployed [1, 2, 3, 4] or datasets that we use to train them [5]. This effort often leads to contributions to ML privacy [6] and security services for cloud infrastructures [7, 8].

Short Bio
I received my B.S. in Electrical Engineering and Computer Science (EECS) from Seoul National University in 2015. During my undergraduate years, I carried out projects with LG Electronics Inc. (LGE) as a lead researcher (as part of my mandatory military service in South Korea). Also, I founded Openwise Inc. (2013-15), a start-up company where I worked as a chief technology officer, and supervised research projects carried out with Samsung Advanced Institute of Technology (SAIT).

Office: [Maryland Cybersecurity Center (MC2)]
5112 Brendan Iribe Center for Computer Science and Engineering, College Park, MD, USA

(New) Jan. 2021: My paper has been accepted to ICLR 2021. (Spotlight Talk)
(New) Nov. 2020: My talk proposal has been accepted to the USENIX Enigma 2021. (Super-excited!)
Oct. 2020: I started my internship at Google Brain under the supervision of Dr. Nicholas Carlini and Dr. Alexey Kurakin.
May. 2020: I did my thesis proposal. Now I became a Ph.D. candidate :)
Mar. 2020: I was awarded the Ann G. Wylie Dissertation Fellowship.
Jan. 2020: I was selected as a 2020-21 Future Faculty Fellow by the Clark School of Engineering.

Selected Publications
— Practical Hardware Attacks on Deep Learning (Project website)
A Panda? No, It's a Sloth: Slowdown Attacks on Adaptive Multi-Exit Neural Network Inference.

*Sanghyun Hong, *Yigitcan Kaya, Ionuţ-Vlad Modoranu, and Tudor Dumitraș. (* indicates equal contribution)
International Conference on Learning Representations, 2021 (ICLR). [Spotlight]
[ Paper ]

How to 0wn NAS in Your Spare Time.

Sanghyun Hong, Michael Davinroy, Yigitcan Kaya, Dana Dachman-Soled, and Tudor Dumitraș.
International Conference on Learning Representations, 2020 (ICLR).
[ Paper | Talk ]

Terminal Brain Damage: Exposing the Graceless Degradation in Deep Neural Networks Under Hardware Fault Attacks.

Sanghyun Hong, Pietro Frigo, Yigitcan Kaya, Cristiano Giuffrida, and Tudor Dumitraș.
USENIX Security Symposium, 2019 (USENIX).
[ Paper | Talk ]

— Practical Data Poisoning Attacks and Defenses
On the Effectiveness of Mitigating Data Poisoning Attacks with Gradient Shaping.

Sanghyun Hong, Varun Chandrasekaran, Yigitcan Kaya, Tudor Dumitraș, and Nicolas Papernot.
arXiv Pre-print 2020.
[ Paper ]

— Turning Distinct Internal Behaviors of DNNs into Defensive Mechanisms
On the Effectiveness of Regularization Against Membership Inference Attacks.

Yigitcan Kaya, Sanghyun Hong, and Tudor Dumitraș.
arXiv Pre-print 2020.
[ Paper ]

Shallow-Deep Networks: Understanding and Mitigating Network Overthinking.

Yigitcan Kaya, Sanghyun Hong, and Tudor Dumitraș.
International Conference on Machine Learning, 2019 (ICML).
[ Paper | Website ]