Towards Robust Multimodal Large Language Models Against Jailbreak Attacks
Ziyi Yin (殷梓译)
Ph.D. Student (4th year) · PSU Data Science Lab
The Penn State University, PA, United States
Email: ziyiyin1998 [at] gmail [dot] com
About me
I am on the job market, seeking a full-time Research Scientist position starting in 2027. If my research aligns with your team's direction, please don't hesitate to reach out via email!
Ziyi Yin is a 4th-year Ph.D. candidate at Penn State University, advised by Prof. Fenglong Ma and Prof. Ting Wang at the PSU Data Science Lab. His research interests lie broadly in data science and artificial intelligence, with a focus on machine learning. His work spans multimodal learning systems, particularly multimodal large language models (MLLMs), with an emphasis on robustness and intrinsic security challenges.
His research also focuses on agentic AI and domain-specific LLMs in health informatics, including the design of multi-agent LLM frameworks for accurate and explainable structured prediction tasks, such as medical coding.
Before that, he obtained his M.S. and B.S. degrees from Xi'an Jiaotong University (Special Class for the Gifted Young, 少年班) in China.
Selected Publications
All publications →Multimodal Robustness and Security
Shadow-Activated Backdoor Attacks on Multimodal Large Language Models
VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models
VLAttack: Multimodal Adversarial Attacks on Vision-Language Tasks via Pretrained Models
Multimodal Applications in Health Informatics
Don't Forget the Past: A History-Augmented Multi-Agent Framework for Explainable ICD Coding
ICDSTAR: Structure-Aware Reasoning for Explainable ICD Coding with Large Language Models
ICDAGENT: Empowering Agentic Large Language Models for Explainable Medical Coding
Asymmetrical Reciprocity-based Federated Learning for Resolving Disparities in Medical Diagnosis
*Equal contribution
Visual Understanding and Object Detection
Multimodal Transformer Network for Pedestrian Trajectory Prediction
Order-independent Matching with Shape Similarity for Parking Slot Detection
Learning to Plan Semantic Free-Space Boundary
Work Experience
Amazon, Inc. (AWS AI Lab)
Summer 2026
Applied Scientist Intern — Active Interaction and Perception Reasoning in VLA Models
Mentors: Sangmin Woo & Kang Zhou
Amazon, Inc. (Store-Foundation AI)
Summer 2025
Applied Scientist Intern — Towards Parameter-Efficiency on SMoE LLMs
Mentors: Zhengyang Wang & Hejie Cui