I am a Ph.D. student in computer science at the University of Massachusetts Amherst, advised by Prof. Amir Houmansadr.

My research focuses on exploring the privacy, security, and trustworthiness of AI models. Recently, I have been investigating the reliability of multimodal models, while also maintaining a strong interest in related areas such as fairness, interpretability, and responsible AI. My current work delves into assessing the trustworthiness of multimodal generative models across diverse domains, including text-to-image and audio-based modalities.

Prior to my graduate studies, I earned my bachelor’s degree in computer engineering from the Hong Kong University of Science and Technology (HKUST) in the year 2023, where I completed my Final Year Thesis (FYT) on the topic of “Adversarial Attacks in Federated Learning” under the supervision of Prof. Jun Zhang. I have also worked with Prof. Minhao Cheng on the robustness of language models, specifically exploring methods associated with backdoor defense in text domain.

---

[Résumé] / [Google Scholar] / [GitHub] / [Linkedin]

📣 News

• Sep 27 ‘24: Our paper “OSLO: One-Shot Label-Only Membership Inference Attacks” was accepted to NeurIPS ‘24! 🎉

• Dec 22 ‘23: Our paper “Memory Triggers: Unveiling Memorization in Text-To-Image Generative Models through Word-Level Duplication” was accepted to the AAAI ‘23 PPAI Workshop! 🎉

📝 Publications

OverThink: Slowdown Attacks on Reasoning LLMs
Abhinav Kumar, Jaechul Roh, Ali Naseh, Marezna Karpinska, Mohit Iyyer, Amir Houmansadr, and Eugene Bagdasaryan
Preprint at arXiv
[paper] [code]

FameBias: Embedding Manipulation Bias Attack in Text-to-Image Models
Jaechul Roh, Andrew Yuan, Jinsong Mao
Preprint at arXiv
[paper]

OSLO: One-Shot Label-Only Membership Inference Attacks
Yuefeng Peng, Jaechul Roh, Subhransu Maji, Amir Houmansadr
NeurIPS 2024
[paper]

Backdooring Bias into Text-to-Image Models
Ali Naseh, Jaechul Roh, Eugene Bagdasaryan, Amir Houmansadr
Preprint at arXiv (Under Review)
[paper] [code]

Memory Triggers: Unveiling Memorization in Text-To-Image Generative Models through Word-Level Duplication
Ali Naseh, Jaechul Roh, Amir Houmansadr
The 5th AAAI Workshop on Privacy-Preserving Artificial Intelligence
[paper]

Understanding (Un)Intended Memorization in Text-to-Image Generative Models
Ali Naseh, Jaechul Roh, Amir Houmansadr
Preprint at arXiv
[paper]

Robust Smart Home Face Recognition under Starving Federated Data
Jaechul Roh, Yajun Fang
Oral Presentation in the IEEE International Conference on Universal Village (IEEE UV2022)
[paper] | [code] | [slides] | [video]

MSDT: Masked Language Model Scoring Defense in Text Domain
Jaechul Roh, Minhao Cheng, Yajun Fang
Oral Presentation in the IEEE International Conference on Universal Village (IEEE UV2022)
[paper] | [code] | [slides] | [video]

Impact of Adversarial Training on the Robustness of Deep Neural Networks
Jaechul Roh
2022 IEEE 5th International Conference on Information Systems and Computer Aided Education (ICISCAE)
[paper] | [code]