Window-based Membership Inference Attacks Against Fine-tuned Large Language Models
基于窗口的细调大语言模型成员推断攻击
机构 * Purdue University(普渡大学) ; Cisco Research(思科研究) ; Cisco Systems(思科系统)
AI总结 本文提出WBC方法,通过滑动窗口和符号聚合技术,有效识别细调大语言模型中的训练数据,显著提升检测性能。
Comments Accepted to USENIX Security 2026. This extended arXiv version includes complete experimental results. The source code is publicly available at: https://github.com/Stry233/WBC/