Zebin Yun

Blavatnik School of Computer Science

Tel Aviv University

Zebin Yun is a Ph.D. student at the Blavatnik School of Computer Science, Tel Aviv University, advised by Dr. Eyal Ronen and Dr. Mahmood Sharif. His research is on the security and privacy of machine learning — including adversarial-example transferability, backdoor attacks on foundation models, and privacy-preserving computation.

Before his Ph.D., he earned his M.Sc. at Tel Aviv University and his B.Eng. in Computer Science at the Southern University of Science and Technology (SUSTech).

I’m always happy to connect and chat, just drop me an email!

news

Jun 15, 2026	Our paper Backdooring Acoustic Foundation Models for Physically Realizable Triggers was accepted to RAID 2026!
Apr 01, 2025	Started my Ph.D. at the Blavatnik School of Computer Science, Tel Aviv University.
Oct 14, 2024	Two papers presented at ACM CCS 2024 workshops: AISec (adversarial-example transferability) and HealthSec (privacy-preserving genomic research).

selected publications

RAID

Backdooring Acoustic Foundation Models for Physically Realizable Triggers

Zebin Yun, Eyal Ronen, and Mahmood Sharif

In International Symposium on Research in Attacks, Intrusions and Defenses (RAID), 2026

Bib

@inproceedings{Yun26FAB,
  title = {Backdooring Acoustic Foundation Models for Physically Realizable Triggers},
  author = {Yun, Zebin and Ronen, Eyal and Sharif, Mahmood},
  booktitle = {International Symposium on Research in Attacks, Intrusions and Defenses (RAID)},
  year = {2026},
}

AISec
The Ultimate Combo: Boosting Adversarial Example Transferability by Composing Data Augmentations

Zebin Yun, Achi-Or Weingarten, Eyal Ronen, and 1 more author

In ACM Workshop on Artificial Intelligence and Security (AISec), 2024

Abs DOI arXiv Bib HTML PDF Code

Deep neural networks are vulnerable to adversarial examples that transfer across models. We systematically study how composing data augmentations affects transferability, exploring 46 augmentation techniques individually and in combination via exhaustive and genetic search. On ImageNet and CIFAR-10 across 18 models, simple color-space augmentations combined with standard ones substantially improve transferability over the state of the art.
@inproceedings{yun2024ultimate, title = {The Ultimate Combo: Boosting Adversarial Example Transferability by Composing Data Augmentations}, author = {Yun, Zebin and Weingarten, Achi-Or and Ronen, Eyal and Sharif, Mahmood}, booktitle = {ACM Workshop on Artificial Intelligence and Security (AISec)}, year = {2024}, doi = {10.1145/3689932.3694769}, }

HealthSec

Privacy-Preserving Collaborative Genomic Research: A Real-Life Deployment and Vision

Zahra Rahmani, Nahal Shahini, Nadav Gat, and 7 more authors

In ACM Workshop on Cybersecurity in Healthcare (HealthSec), 2024

DOI arXiv Bib HTML

@inproceedings{rahmani2024privacy,
  title = {Privacy-Preserving Collaborative Genomic Research: A Real-Life Deployment and Vision},
  author = {Rahmani, Zahra and Shahini, Nahal and Gat, Nadav and Yun, Zebin and Jiang, Yuzhou and Farchy, Ofir and Harel, Yaniv and Chaudhary, Vipin and Ayday, Erman and Sharif, Mahmood},
  booktitle = {ACM Workshop on Cybersecurity in Healthcare (HealthSec)},
  year = {2024},
  doi = {10.1145/3689942.3694747},
}