👋 Hey there! I’m Jing Shao.
I’m a graduate student in Computer Software Engineering at Northeastern University – College of Engineering, specializing in machine learning, multi-modal model safety, and end-to-end AI system development.
My research focuses on making generative and embodied AI systems more trustworthy, robust, and efficient, aiming for reliable performance in real-world scenarios.
Beyond academia, I contributed to a stealth-stage AI startup, developing AI-powered audio systems and integrating ML models into scalable infrastructure for real-time interaction. I’m also a Google Summer of Code (GSoC) 2025 contributor, working on secure Flutter engine integration for the open-source community.
📌 I’m actively seeking Research Assistant and PhD position.
✨ My CV can be found here! Feel free to reach out or connect!!
🔥 News
- Aug 2025 — 🎉 Selected as a Mentor for AWS re:Invent 2025, mentoring 10 ABW Grant participants.
- May 2025 — 🎉 Awarded the AWS re:Inforce 2025 Grant for my work in AI safety research.
- May 2025 — 🎉 Selected as a Google Summer of Code 2025 Contributor.
📝 Publications
Hao Cheng, Erjia Xiao, Jing Shao, Yichi Wang, Le Yang, Chao Shen, Philip Torr, Jindong Gu, Renjing Xu
- Academic Impact: This work introduces Jailbreak-AudioBench, the first comprehensive benchmark for evaluating audio-based jailbreak threats against LALMs. By systematically exploiting audio-specific hidden semantics—such as tone, intonation, background noise, and emotion—we uncover previously unreported vulnerabilities in multimodal AI safety. This research lays the foundation for future work in understanding and defending against audio manipulation attacks in LALMs.
- Practical Impact: Our benchmark reveals critical security vulnerabilities in state-of-the-art LALMs, with models like SALMONN-7B reaching an 85.1% Attack Success Rate under systematic query-based attacks. These findings have direct implications for real-world systems, including voice assistants, customer service bots, and in-vehicle audio interfaces, offering actionable guidance for building safer and more robust audio AI systems.
Energy-based Active Learning for Bringing Beam-induced Domain Gap for 3D Object Detection
Le Yang, Yixuan Yan, Jing Shao, Hao Cheng, Fan Li
- Academic Impact: This work proposes an energy-based active learning framework to close the domain gap between 64-beam and 16-beam LiDARs, enabling knowledge transfer with minimal labeled data.
- Practical Impact: Our method significantly reduces annotation cost and achieves high-performance 3D object detection on 16-beam LiDAR with only a small portion of labeled samples.
🎖 Honors and Awards
- May 2025, Awarded AWS All Builders Welcome Grant re:Inforce.
- May 2025, GSoC 2025 Contributor & Flutter Organization Member.
- Sep 2024, Awarded AWS All Builders Welcome Grant re:Invent.
📖 Educations
- Jan 2024 – Apr 2026, M.S., Computer Software Engineering, Northeastern University.
📣 Talks and Conferences
- Jun 2025, Attending AWS re:Inforce 2025, Philadelphia.
- Dec 2024, Attending AWS re:Invent 2024, Las Vegas.
- Oct 2024, Presenting research at MobiCom 2024, Washington D.C.
💻 Internships
- May 2025 – Present, Google Summer of Code, Software Developer.
- Jan 2025 – Present, Stealth AI Startup, Machine Learning Engineer.
- Sep 2024 – May 2025, The Hong Kong University of Science and Technology, Research Assistant.
- May 2024 – Sep 2024, X-Humanoid, Machine Learning Engineer.
📝 Blogs
I write about my journey in tech and my thoughts in life!