Wei Song

Lecturer (~U.S. Assistant Professor), School of Information and Communication Technology (ICT), Griffith University

Google Scholar | LinkedIn | w.song@griffith.edu.au
Wei Song

📢 PhD Students:
I am a Lecturer at Griffith University, and I am looking for self-motivated PhD students to build a research group in AI security — adversarial machine learning, AI watermarking, and AI security for energy systems. Please email me with your CV and transcript if your research interests overlap with mine.

Master's, honours, and capstone students: I am happy to discuss thesis projects linked to my funded industry projects in secure AI and content authenticity. I currently co-mentor three PhD and three Master's students across UNSW Sydney, the National University of Singapore, and Fuzhou University.

Biography

I am a Lecturer (~U.S. Assistant Professor) in the School of Information and Communication Technology (ICT) at Griffith University. Before that, I was a Postdoctoral Research Fellow in the School of Computer Science and Engineering (CSE) at UNSW Sydney, working with Prof. Jingling Xue and Prof. Yulei Sui. I received my Ph.D. in Computer Science and Engineering from UNSW Sydney in 2024, supervised by Prof. Jingling Xue.

My research interests focus on AI security, with particular emphasis on adversarial machine learning, AI watermarking, and AI security for energy systems. My work has been published at leading venues, including USENIX Security, RAID, AAAI, ASE, ACM UbiComp/IMWUT, ICME, and IEEE IoTJ. I have secured AUD 3.7M funding for AI security in energy systems. I am also a co-founder of SunstackAI (Spinout), which provides trustworthy, energy-aware computing load balancing for a distributed GPU network.

News

  • 08/2026Our paper rethinking MCP-augmented LLMs has been accepted by EMNLP 2026.
  • 08/2026🎉 I will be joining the School of Information and Communication Technology at Griffith University as a Lecturer (~U.S. Assistant Professor). I am recruiting PhD students — see above.
  • 06/2026Our TRaCE Lab to Market project “Secure Federated Battery Management System for Distributed Energy Storage Assets” (AUD 1.3M) is now active. I serve as Lead CI.
  • 04/2026Two papers on dataset poisoning and LLM debiasing have been accepted by ACL Findings 2026.
  • 03/2026Our paper on adversarial attacks against voice control systems has been accepted by ICME 2026.
  • 02/2026Our paper on membership inference against Video LLMs has been accepted by SiMLA 2026.
  • 01/2026Our demo paper on CAPTCHA for modern web applications has been accepted by WWW 2026.
  • 11/2025Our paper on failures to surface harmful content in Video LLMs has been accepted by AAAI 2026.
  • 07/2025Our paper on adversarial video defense (ViDToken) has been accepted by RAID 2025.
  • 06/2025Our paper on hand biometric authentication has been accepted by IMWUT/UbiComp 2025.
  • 05/2025Our paper on adversarial defense for point cloud classification has been accepted by IEEE ICRA 2025.
  • 2025VibMilk progressed to a UNSW Invention Record (D2025-0023), with 20+ media reports including ABC Landline and New Atlas.
  • 11/2024Started as a Postdoctoral Research Fellow at UNSW Sydney.
  • 06/2024My first work on adversarial video defense (SecVID) has been accepted by USENIX Security 2024.
  • 03/2024Our paper “VibMilk: Non-Intrusive Milk Spoilage Detection via Smartphone Vibration” has been accepted by IEEE IoTJ.
  • 01/2024Our paper “CAvatar: Real-time Human Activity Mesh Reconstruction via Tactile Carpets” has been accepted by IMWUT/UbiComp 2024.

Research Directions

  • Adversarial Machine Learning
  • AI Watermarking
  • AI Security for Energy Systems
    • Privacy-preserving federated learning for distributed battery health [TRaCE, Lead CI, 1.3M AUD]
    • BMS software-interface security and residential AI compute [Theme Lead CI, 2.4M AUD]
    • Secure communication and telemetry data integrity [RACE, Theme Lead CI, 2.0M AUD]

Grants & Funding

Trailblazer Recycling and Clean Energy (TRaCE) RACE for 2030 Australian Renewable Energy Agency (ARENA)
  • Lead CI in the project “Secure Federated Battery Management System for Distributed Energy Storage Assets”, TRaCE, 1,300,000 AUD (cash). Awarded
  • Theme Lead CI in the project “Trustworthy, Energy-Aware Computing Load Balancing for a Distributed Residential GPU Network”, 2,400,000 AUD (cash). Awarded

Publications

(# Equal contribution,  * Corresponding author)

2026 (8)
  • Help or Hurdle? Rethinking Model Context Protocol-Augmented Large Language Models CORE A*
    Wei Song, Haonan Zhong, Ziqi Ding, Zhenchang Xing, Yulei Sui, Jingling Xue, Yuekang Li.
    EMNLP '26 PDF
  • RamFuzz: LLM-Guided Greybox Fuzzing for Spatial Memory Corruption via Valid Range Violation CORE A*
    Shangzhi Xu, Wei Song, Yuekang Li, Nan Sun, Muhammad Ejaz Ahmed, Willy Susilo, Benjamin Turnbull, Xiao Cheng*, Siqi Ma*.
    ASE '26
  • NgCaptcha: A CAPTCHA Bridging the Past and the Future
    Ziqi Ding, Shangzhi Xu, Wei Song, Yuekang Li.
    WWW '26 (Demo) PDF
  • Membership Inference Attacks Against Video Large Language Models
    Wei Song, Yuxin Cao, Ziqi Ding, Yi Liu, Gelei Deng, Yuekang Li.
    SiMLA '26
  • DUAP: Dual-task Universal Adversarial Perturbations Against Voice Control Systems CORE A
    Suyang Sun, Weifei Jin, Yuxin Cao, Wei Song, Jie Hao.
    ICME '26 PDF
  • Debiasing LLMs by Masking Unfairness-Driving Attention Heads CORE A*
    Tingxu Han, Wei Song, Ziqi Ding, Ziming Li, Chunrong Fang, Yuekang Li, Dongfang Liu, Zhenyu Chen, Zhenting Wang.
    ACL Findings '26 PDF
  • Train in Vain: Functionality-Preserving Poisoning to Prevent Unauthorized Use of Code Datasets CORE A*
    Yuan Xiao, Yuchen Chen, Jiaming Wang, Wei Song, Jun Sun, Shiqing Ma, Yanzhou Mu, Juan Zhai, Chunrong Fang, Jin Song Dong, Zhenyu Chen.
    ACL Findings '26
  • Failures to Surface Harmful Contents in Video Large Language Models CORE A*
    Yuxin Cao, Wei Song, Derui Wang, Jingling Xue, Jin Song Dong.
    AAAI '26 PDF
2025 (4)
  • HandID: Unobtrusive Gesture-independent User Authentication on Smartphones Using Vibration-based Hand Biometrics CORE A*
    Yuezhong Wu#, Wei Song#, Chun Tung Chou, Jiankun Hu, Wen Hu.
    ACM IMWUT/UbiComp '25 PDF
  • ViDToken: A Video-Transformer-Based Latent Token Defense for Adversarial Video Detection CORE A
    Wei Song, Zhenchang Xing, Liming Zhu, Yulei Sui, Jingling Xue.
    RAID '25 PDF
  • ai.txt: A Domain-Specific Language for Guiding AI Interactions with the Internet
    Yuekang Li, Wei Song, Bangshuo Zhu, Dong Gong, Yi Liu, Gelei Deng, Chunyang Chen, Lei Ma, Jun Sun, Toby Walsh, Jingling Xue.
    arXiv:2505.07834 PDF
  • SynerGuard: A Robust Framework for Point Cloud Classification via Local Geometry and Spatial Topology CORE A
    Haonan Zhong, Wei Song, Maurice Pagnucco, Yang Song.
    IEEE ICRA '25 PDF
2024 (4)
  • Can Distillation Mitigate Backdoor Attacks in Pre-trained Encoders?
    Tingxu Han, Wei Song, Weisong Sun, Ziqi Ding, Yebo Feng, Chunrong Fang, Jun Li, Hanwei Qian, Zhenyu Chen, Yang Liu.
    arXiv:2403.03846 PDF
  • VibMilk: Non-Intrusive Milk Spoilage Detection via Smartphone Vibration JCR Q1 · IF 9.0 UNSW Invention Record
    Yuezhong Wu, Wei Song, Yanxiang Wang, Dong Ma, Weitao Xu, Mahbub Hassan, Wen Hu.
    IEEE Internet of Things Journal '24 PDF
  • CAvatar: Real-Time Human Activity Mesh Reconstruction via Tactile Carpets CORE A*
    Wenqiang Chen, Yexin Hu, Wei Song, Yingcheng Liu, Antonio Torralba, Wojciech Matusik.
    ACM IMWUT/UbiComp '24 PDF
  • SecVID: Correction-based Defense Against Adversarial Video Attacks via Discretization-Enhanced Video Compressive Sensing CORE A*
    Wei Song, Cong Cong, Haonan Zhong, Jingling Xue.
    USENIX Security '24 PDF
2022 (3)
  • Pistis: Replay Attack and Liveness Detection for Gait-Based User Authentication on Wearable Devices Using Vibration JCR Q1 · IF 9.0
    Wei Song, Hong Jia, Min Wang, Yuezhong Wu, Wanli Xue, Chun Tung Chou, Jiankun Hu, Wen Hu.
    IEEE Internet of Things Journal '22 PDF
  • Towards Behavior-Independent In-Hand User Authentication on Smartphone Using Vibration
    Wei Song, Min Wang, Yuezhong Wu, Chun Tung Chou, Jiankun Hu, Wen Hu.
    ACM MobiCom '22 (Poster) PDF
  • Soil Moisture Sensing with mmWave Radar
    Wenyao Chen, Yimeng Feng, Mark Cardamis, Cheng Jiang, Wei Song, Oula Ghannoum, Wen Hu.
    ACM mmNets '22 (MobiCom Workshop) PDF
2021 (1)
  • Towards Privacy-Preserving Digital Contact Tracing
    Nadeem Ahmed, Regio A. Michelin, Wanli Xue, Guntur Dharma Putra, Wei Song, Sushmita Ruj, Salil S. Kanhere, Sanjay Jha.
    IEEE ICBC '21 PDF
Preprints & Under Review (7)
  • DeMark: A Query-Free Black-Box Attack on Deepfake Watermarking Defenses
    Wei Song, Zhenchang Xing, Liming Zhu, Yulei Sui, Jingling Xue.
    Under review, IEEE TIFS, 2026 PDF
  • Enhancing and Reporting Robustness Boundary of Neural Code Models for Intelligent Code Understanding
    Tingxu Han, Wei Song, Weisong Sun, Hao Wu, Chunrong Fang, Yuan Xiao, Xiaofang Zhang, Zhenyu Chen, Yang Liu.
    arXiv:2603.24119, 2026 PDF
  • SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?
    Tingxu Han, Yi Zhang, Wei Song, Chunrong Fang, Zhenyu Chen, Youcheng Sun, Lijie Hu.
    arXiv:2603.15401, 2026 PDF
  • FAIRT2V: Training-Free Debiasing for Text-to-Video Diffusion Models
    Haonan Zhong, Wei Song, Tingxu Han, Maurice Pagnucco, Jingling Xue, Yang Song.
    arXiv:2601.20791, 2026 PDF
  • VideoSTF: Stress-Testing Output Repetition in Video Large Language Models
    Yuxin Cao, Wei Song*, Shangzhi Xu, Jingling Xue, Jin Song Dong.
    arXiv:2602.10639, 2026 PDF
  • Poisoning Prompt-Guided Sampling in Video Large Language Models
    Yuxin Cao, Wei Song, Jingling Xue, Jin Song Dong.
    arXiv:2509.20851 PDF
  • Robust CAPTCHA Using Audio Illusions in the Era of Large Language Models: from Evaluation to Advances
    Ziqi Ding, Yunfeng Wang, Wei Song, Yi Liu, Gelei Deng, Nan Sun, Huadong Mo, Jingling Xue, Shidong Pan, Yuekang Li.
    arXiv:2601.08516 PDF

Supervision & Mentorship

PhD Students (3)
  • Yuxin Cao (PhD at National University of Singapore) — safety evaluation of Video LLMs.
  • Bangshuo Zhu (PhD at UNSW Sydney) — robust and trustworthy defense methods for Video LLMs.
  • Haonan Zhong (PhD at UNSW Sydney) — safety alignment of large language models for trustworthy deployment.
Master's Students (3)
  • Ziqi Ding (UNSW Sydney) — next-generation CAPTCHA design for secure and usable authentication.
  • Yingbo Xie (Fuzhou University) — adversarial attacks and defenses for robust gait recognition.
  • Peng Yang (Fuzhou University) — reliability and failure analysis of Video LLMs.

Professional Service

  • Artifact Evaluation Committee, USENIX Security.
  • Reviewer for AI and machine learning venues: AAAI, ICLR, ICML, CVPR, ICME.
  • Reviewer for software engineering venues: IEEE/ACM ASE, FSE.
  • Reviewer for systems and networking venues: ACM SIGCOMM, ACM MobiSys, ACM SenSys, IEEE IoTDI, ACM/IEEE IPSN.

Education & Experience

  • 2026 – Present, Lecturer (~U.S. Assistant Professor), School of Information and Communication Technology, Griffith University.
  • Nov 2024 – 2026, Postdoctoral Research Fellow, UNSW Sydney.
  • 2019 – 2024, Tutor / Course Administrator, School of Computer Science and Engineering, UNSW Sydney.
  • Jan 2022 – May 2022, Software Engineer, Parking Spotz. Mentor: Prof. Wen Hu.
  • Dec 2020 – May 2021, Research Assistant, UNSW / Cyber Security Cooperative Research Centre. Privacy-preserving mobile technologies for COVID-19 contact tracing; received UNSW News coverage and was incorporated into COMP4337/9337 teaching materials. Supervisors: Prof. Salil Kanhere and Prof. Sanjay Jha.
  • Mar 2021 – Nov 2024, Ph.D. in Computer Science and Engineering, UNSW Sydney. Supervisor: Prof. Jingling Xue.
  • Mar 2019 – Dec 2020, Master of Information Technology, UNSW Sydney.
  • Sep 2014 – Jun 2018, Bachelor of Civil Engineering, Southwest Jiaotong University (SWJTU), China.