Samra Irshad

samra@khu.ac.kr

I am a 4th-year PhD candidate at Kyung Hee University, South Korea. My research focuses on identifying and modeling novel, real-world-inspired scenarios in which deployed AI systems could be vulnerable to adversarial manipulation. In particular, I am interested in understanding the security implications of deploying Large Language Models (LLMs), Multimodal Large Language Models (MLLMs), and generative models (GANs/Diffusion), and in developing adversarial threats that exploit the attack surfaces introduced by their deployment.

Previously, I worked at See-Mode Technologies as a Machine Learning Engineer, where I contributed to the development of an automated thyroid ultrasound reporting system. My work focused on preprocessing large ultrasound datasets and designing and implementing deep learning algorithms to detect thyroid nodules in ultrasound images.

Selected Publications

MIDL

No Evidence of Disease: Clinically-Risky Adversarial Chest CT Report Generation

Samra Irshad, Junho Kim, and Seong Tae Kim

In Proceedings of the Medical Imaging with Deep Learning (MIDL), 2026

Bib HTML

@inproceedings{irshad2026noevidence,
  title = {No Evidence of Disease: Clinically-Risky Adversarial Chest CT Report Generation},
  author = {Irshad, Samra and Kim, Junho and Kim, Seong Tae},
  booktitle = {Proceedings of the Medical Imaging with Deep Learning (MIDL)},
  year = {2026},
  address = {Taipei, Taiwan},
}

IEEE TDSC

Adversarial Wear and Tear: Exploiting Natural Damage for Generating Physical-World Adversarial Examples

Samra Irshad, Seungkyu Lee, Nassir Navab, and 2 more authors

IEEE Transactions on Dependable and Secure Computing, 2026

DOI arXiv Bib Code

@article{irshad2026adversarial,
  title = {Adversarial Wear and Tear: Exploiting Natural Damage for Generating Physical-World Adversarial Examples},
  author = {Irshad, Samra and Lee, Seungkyu and Navab, Nassir and Lee, Hong Joo and Kim, Seong Tae},
  journal = {IEEE Transactions on Dependable and Secure Computing},
  year = {2026},
  publisher = {IEEE},
  doi = {10.1109/TDSC.2026.3660107},
}

arXiv

SegSTRONG-C: Segmenting Surgical Tools Robustly On Non-adversarial Generated Corruptions–An EndoVis’ 24 Challenge

Hao Ding, Yuqian Zhang, Tuxun Lu, and 9 more authors

arXiv preprint arXiv:2407.11906, 2024

DOI arXiv Bib

@article{ding2024segstrong,
  title = {SegSTRONG-C: Segmenting Surgical Tools Robustly On Non-adversarial Generated Corruptions--An EndoVis' 24 Challenge},
  author = {Ding, Hao and Zhang, Yuqian and Lu, Tuxun and Liang, Ruixing and Shu, Hongchao and Seenivasan, Lalithkumar and Long, Yonghao and Dou, Qi and Gao, Cong and Leng, Yicheng and Irshad, Samra and others},
  journal = {arXiv preprint arXiv:2407.11906},
  year = {2024},
  doi = {https://doi.org/10.48550/arXiv.2407.11906},
}

IEEE Access

Improved abdominal multi-organ segmentation via 3d boundary-constrained deep neural networks

Samra Irshad, Douglas PS Gomes, and Seong Tae Kim

IEEE Access, 2023

DOI Bib Code

@article{irshad2023improved,
  title = {Improved abdominal multi-organ segmentation via 3d boundary-constrained deep neural networks},
  author = {Irshad, Samra and Gomes, Douglas PS and Kim, Seong Tae},
  journal = {IEEE Access},
  volume = {11},
  pages = {35097--35110},
  year = {2023},
  publisher = {IEEE},
  doi = {10.1109/ACCESS.2023.3264582},
}

Professional Activities

Conference Reviewer

MICCAI Main Conference 2020, 2022 (Outstanding Reviewer Honourable Mention), 2023, 2025
MICCAI ASMUS Workshop 2024 (International Workshop on Advances in Simplifying Medical Ultrasound)
MICCAI Student Board EMERGE Workshop 2024
NeurIPS Main Conference 2024
NeurIPS MINT Workshop 2024 (Workshop on Foundation Model Interventions)
NeurIPS AIM-FM Workshop 2024 (Advancements In Medical Foundation Models: Explainability, Robustness, Security, and Beyond)
IJCAI Workshop 2021 (Workshop on Weakly Supervised Representation Learning)

Journal Reviewer

IEEE Transactions on Medical Imaging
Computer Methods and Programs in Biomedicine
IEEE Access
Computers in Biology and Medicine
Computerized medical imaging and graphics