Skip to content

about.md

Latest commit

 

History

History
49 lines (40 loc) · 3.55 KB

about.md

File metadata and controls

49 lines (40 loc) · 3.55 KB
permalink /
title
excerpt
author_profile true
redirect_from
/about/
/about.html

{% if site.google_scholar_stats_use_cdn %} {% assign gsDataBaseUrl = "https://cdn.jsdelivr.net/gh/" | append: site.repository | append: "@" %} {% else %} {% assign gsDataBaseUrl = "https://raw.githubusercontent.com/" | append: site.repository | append: "/" %} {% endif %} {% assign url = gsDataBaseUrl | append: "google-scholar-stats/gs_data_shieldsio.json" %}

Welcome!!!😘😘😘

I'm Yijie Lu (陆一杰)🐕. I am an undergraduate student at the School of Cyber Science and Engineering, Wuhan University (WHU), majoring in Cyberspace Security. I will be an incoming Ph.D. student at the School of Computer Science, Shanghai Jiao Tong University (SJTU), starting in 2026, under the supervision of Prof. Zhuosheng Zhang.

My current research interests include AI Security and GUI Agent. I am actively seeking collaborations and look forward to connecting with more people. If you are interested in my work, please drop me an email!☀️

📝 Publications & Research

  • EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks [arXiv]
    Accepted to ACL 2026 Findings.
    This paper introduces EVA, an evolutionary framework designed to red-team GUI agents against Environmental Injection Attacks (EIAs) by evolving adversarial payloads exclusively within the semantic dimension. Guided by controlled experiments demonstrating that semantic deception—rather than visual appearance—is the primary determinant of attack success, the framework mines linguistic vulnerabilities and distills them into generalizable rules. Empirical results show that EVA achieves up to an 85% attack success rate across five representative victim agents, evolving benign seeds into successful attacks in only 1.18 to 1.71 iterations. This rapid convergence exposes a dense semantic attack space and an "alignment paradox," revealing that instruction-following capabilities reinforced by alignment training can inadvertently render agents inherently susceptible to authoritative, semantically deceptive environmental cues.

🎖 Scholarships and Honors

  • Outstanding Graduate of WHU (Top 10%) Wuhan University
  • Lei Jun Computer Science Undergraduate Scholarship Wuhan University & Xiaomi Inc.
  • First Class Scholarship of WHU (Top 5%) Wuhan University
  • Merit Student (Top 10%) Wuhan University
  • LvMeng Scholarship Wuhan University
  • Advanced Individual in Scientific and Technological Innovation Wuhan University

📖 Educations

  • 2026.09 - present, P.hD. in Cyberspace Security, Shanghai Jiao Tong University, China.
  • 2022.08 - 2026.06, B.E. in Cyberspace Security, Wuhan University, China.

🎡 Activities & Services

  • Teaching Assistant, WHU-Jisuanke Joint Practical Training Course "Security Maker Practice", Jun 2025 - Jul 2025
    • Awarded "Top TA" for contributions to teaching, exercise explanation, and Q&A sessions.
  • Teaching Assistant, WHU-Jisuanke Joint Practical Training Course "Security Maker Practice", Jul 2024 - Aug 2024
    • Awarded "Excellent TA" for contributions to the one-month practical training course.
  • Committee Member, Study Committee of the Class 3, Grade 2022, Aug 2022 - Present
  • Member, Front-end Group, Ziqiang Student Network Culture Studio, Aug 2022 - Jun 2023