Founder · AI Agents · ML Security
I'm the founder of Trenz, building AI agents for social commerce in Sunnyvale, CA.
Before starting Trenz in 2024, I was a Research Scientist at ByteDance AI Lab and a Senior Security Scientist at Baidu X-Lab, working on adversarial machine learning and the security of autonomous vehicles, smart homes and mobile apps.
I received my Ph.D. in Computer Science and Engineering from the University of Michigan, advised by Prof. Z. Morley Mao, and my B.E. in Software Engineering from Shanghai Jiao Tong University, advised by Prof. Haibo Chen.
Perceptron Benchmark is a toolbox for measuring the safety and security of deep neural networks. It estimates a model's true robustness from both sides: tightening the empirical upper bound with state-of-the-art attacks, and raising the theoretical lower bound with linear-approximation-based verification. It is framework-agnostic and supports classification and detection models on PyTorch, TensorFlow, Keras and cloud APIs.
Attacks on self-driving and ML-as-a-service pipelines: physical adversarial patches that fool object detectors, fingerprinting attacks against commercial MLaaS APIs, adversarial examples that transfer across real-world vision tasks, and the first adversarial attack against multiple object tracking.
Bit-flipping fuzzers such as AFL and libFuzzer struggle when inputs are large and highly structured. Libprotobuf-mutator for Apollo learns the input format from protocol buffers and generates spec-conforming inputs, reaching deep program logic in the Baidu Apollo autonomous-driving platform. It is now integrated upstream.
OpenCAV / AVGuard. As U-M opened its autonomous vehicle to third-party developers, we built static and dynamic vetting for self-driving apps that may be flawed, vulnerable or malicious. TrafficNet is a large-scale, extensible library of naturalistic driving scenarios that turns research datasets into something vehicle engineers can actually use.
Open ports in Android apps. A static analysis tool that found 400+ apps with exploitable open ports, enabling remote theft of contacts and photos, malware installation and code execution — including session hijacking of AirDroid (CVE-2016-5227).
Covered by WIRED, TechRepublic, Mashable, Bleeping Computer, Axios, Digital Trends and others.
ContexIoT. App patching that brings contextual integrity to commodity smart-home apps with minimal overhead. SInspector. Exposes misuse of Unix domain sockets in popular Android apps and system daemons, including a privilege escalation in KingRoot.
Dirty Road Can Attack: Security of Deep Learning based Automated Lane Centering under Physical-World Attack
Takami Sato, Junjie Shen, Ningfei Wang, Yunhan Jia, Xue Lin, Qi Alfred Chen
USENIX Sec'21USENIX Security Symposium
Enhancing Cross-Task Black-Box Transferability of Adversarial Examples with Dispersion Reduction
Yantao Lu, Yunhan Jia, Jianyu Wang, Bai Li, Weiheng Chai, Lawrence Carin, Senem Velipasalar
CVPR'20IEEE/CVF Conference on Computer Vision and Pattern Recognition
Explainable Machine Learning in Deployment
Umang Bhatt, Alice Xiang, Shubham Sharma, Adrian Weller, Ankur Taly, Yunhan Jia, Joydeep Ghosh, Ruchir Puri, José M. F. Moura, Peter Eckersley
FAT*'20ACM Conference on Fairness, Accountability, and Transparency
Fooling Detection Alone is Not Enough: Adversarial Attack against Multiple Object Tracking
Yunhan Jia, Yantao Lu, Junjie Shen, Qi Alfred Chen, Hao Chen, Zhenyu Zhong, Tao Wei
ICLR'20International Conference on Learning Representations
Towards Practical Lottery Ticket Hypothesis for Adversarial Training
Bai Li, Shiqi Wang, Yunhan Jia, Yantao Lu, Zhenyu Zhong, Lawrence Carin, Suman Jana
arXivPreprint
Fooling Detection Alone is Not Enough: First Adversarial Attack against Multiple Object Tracking
Yunhan Jia, Yantao Lu, Junjie Shen, Qi Alfred Chen, Zhenyu Zhong, Tao Wei
AdvMLCV'19CVPR Workshop · Contributed talk
Enhancing Cross-task Transferability of Adversarial Examples with Dispersion Reduction
Yunhan Jia*, Yantao Lu*, Senem Velipasalar, Zhenyu Zhong, Tao Wei
AdvMLCV'19CVPR Workshop
Attacking Multiple Object Tracking using Adversarial Examples
Yunhan Jia, Yantao Lu, Junjie Shen, Qi Alfred Chen, Zhenyu Zhong, Tao Wei
SPML'19ICML Workshop on Security and Privacy of Machine Learning
The Cost of Learning from the Best: How Prior Knowledge Weakens the Security of Deep Neural Networks
Yunhan Jia, Zhenyu Zhong, Yulong Zhang, Qian Feng, Tao Wei
BH Asia'19Black Hat Asia, Singapore
Perception Deception: Physical Adversarial Attack Challenges and Tactics for DNN-based Object Detection
Yunhan Jia, Zhenyu Zhong, Weilin Xu, Tao Wei
BH EU'18Black Hat Europe, London
Dissecting HAS VOD Services for Cellular: Performance, Root Causes and Best Practices
Shichang Xu, Yunhan Jia, Z. Morley Mao, Subhabrata Sen
IMC'17ACM Internet Measurement Conference
TrafficNet: An Open Naturalistic Driving Scenario Library
Ding Zhao, Yaohui Guo, Yunhan Jia
ITSC'17IEEE Intelligent Transportation Systems Conference
Towards Secure and Safe Appified Automated Vehicles
Yunhan Jia, Ding Zhao, Qi Alfred Chen, Z. Morley Mao
IV'17IEEE Intelligent Vehicles Symposium
ContexIoT: Towards Providing Contextual Integrity to Appified IoT Platforms
Yunhan Jia, Qi Alfred Chen, Shiqi Wang, Amir Rahmati, Earlence Fernandes, Z. Morley Mao, Atul Prakash
NDSS'17Network and Distributed System Security Symposium · 16.0% acceptance
Open Doors for Bob and Mallory: Open Port Usage in Android Apps and Security Implications
Yunhan Jia, Qi Alfred Chen, Yikai Lin, Chao Kong, Z. Morley Mao
EuroS&P'17IEEE European Symposium on Security and Privacy · 19.8% acceptance
The Misuse of Android Unix Domain Sockets and Security Implications
Yuru Shao, Jason Ott, Yunhan Jia, Zhiyun Qian, Z. Morley Mao
CCS'16ACM Conference on Computer and Communications Security · 16.5% acceptance
Static Detection of Packet Injection Vulnerabilities: A Case for Identifying Attacker-controlled Implicit Information Leaks
Qi Alfred Chen, Zhiyun Qian, Yunhan Jia, Yuru Shao, Z. Morley Mao
CCS'15ACM Conference on Computer and Communications Security · 19.8% acceptance
Performance Characterization and Call Reliability Problem Diagnosis for Voice over LTE
Yunhan Jia, Qi Alfred Chen, Z. Morley Mao, Jie Hui, Kranthi Sontineni, Alex Yoon, Samson Kwong, Kevin Lau
MobiCom'15ACM International Conference on Mobile Computing and Networking · 18.4% acceptance
Patents
Experience
Education
Service
Honors
Classic rock has had me since 7th grade, and I've played guitar in bands ever since. Favourite band of all time: Guns N' Roses; the boss of vocals will always be Steven Tyler. A few memories of jamming with friends: junior, senior, graduation, grad school.
Hometown Xinjiang. Favourite place so far: Cefalù, Sicily.
