man3kin3ko/awesome-adversarial-machine-learning
GitHub: man3kin3ko/awesome-adversarial-machine-learning
一份精心整理的对抗性机器学习攻击与防御资源列表,涵盖威胁建模、攻防框架、案例研究和CTF竞赛等核心内容。
Stars: 10 | Forks: 4
# 极佳的对抗性机器学习 [](https://awesome.re)
精选的绝佳机器学习安全资源列表。
另请参阅 [awesome-ml-for-cybersecurity](https://github.com/jivoi/awesome-ml-for-cybersecurity) 和 [权威的安全数据科学与机器学习指南](http://www.covert.io/the-definitive-security-datascience-and-machinelearning-guide/)。
- [极佳的对抗性机器学习](#awesome-adversarial-machine-learning-)
- [术语](#terminology)
- [威胁建模](#threat-modeling)
- [控制指南](#controls-guidelines)
- [案例研究](#case-studies)
- [基于领域的攻击](#attacks-based-on-domain)
- [基于策略的攻击](#attacks-based-on-strategy)
- [CTF 和黑客马拉松](#ctf-and-hackathons)
- [框架](#frameworks)
## 术语
* [NIST:对抗性机器学习的分类和术语](https://nvlpubs.nist.gov/nistpubs/ir/2019/NIST.IR.8269-draft.pdf)
## 威胁建模
* [ENISA:人工智能网络安全挑战](https://www.enisa.europa.eu/news/publications/artificial-intelligence-cybersecurity-challenges)
* [MITRE:人工智能系统的对抗性威胁图谱](https://atlas.mitre.org/)
* [攻击性 AI 对组织的威胁](https://arxiv.org/pdf/2106.15764.pdf)
* [计算机视觉中深度学习的对抗性攻击威胁:综述](https://arxiv.org/abs/1801.00553)
## 控制指南
* [ENISA:保护机器学习算法安全](https://www.enisa.europa.eu/publications/securing-machine-learning-algorithms)
* [AISecOps](https://github.com/oasiszrz/awesome-AISecOps)
## 案例研究
* [MITRE 关于现实环境中的报告](https://github.com/mitre/advmlthreatmatrix/blob/master/pages/case-studies-page.md#case-studies-page)
* [Avito 利用对抗性攻击对抗内容窃取](https://habr.com/ru/company/avito/blog/452142/)
* [芝加哥大学研究人员的 Project Nightshade](https://nightshade.cs.uchicago.edu/whatis.html),帮助数字艺术家保护其作品不被用作训练数据。关键的攻击方法是 poisoning。
* [芝加哥大学研究人员的 Project Glaze](https://glaze.cs.uchicago.edu/faq.html),类似于 Nightshade,但通过 mimicry 攻击起作用。
## 基于领域的攻击
* 计算机视觉
* 语音识别
* 特定模型研究
* [Kaldi](https://github.com/lealeasch/adversarialattacks)
* [Lingvo](https://github.com/yaq007/cleverhans/tree/master/examples/adversarial_asr)
* [Deepspeech](https://arxiv.org/pdf/1801.01944)
* 方法
* [电梯内中间人攻击](https://www.usenix.org/sites/default/files/conference/protected-files/woot15_slides_vaidya.pdf)
* 噪声隐藏技术
* [DolphinAttack](https://github.com/USSLab/DolphinAttack)
* [MPEG 压缩](https://arxiv.org/pdf/1808.05665)
## 基于策略的攻击
* 信息收集
* [成员推断](https://arxiv.org/pdf/1610.05820)
* [去匿名化](https://www.cs.utexas.edu/~shmat/shmat_oak08netflix.pdf)
* [模型求逆](https://dl.acm.org/doi/10.1145/2810103.2813677)
* [模型窃取](https://arxiv.org/pdf/1805.02628)
* [盲点检测](https://arxiv.org/pdf/1901.04684)
* [状态预测](https://ieeexplore.ieee.org/document/8716085)
* 拒绝服务
* [Poisoning DoS](https://arxiv.org/pdf/1708.08689.pdf)
* [Sponge 示例](https://arxiv.org/pdf/2006.03463)
* 生物特征欺骗
* [主指纹](https://arxiv.org/pdf/1705.07386)
* [人脸识别逃逸](https://dl.acm.org/doi/10.1145/2976749.2978392)
## CTF 和黑客马拉松
* [NIPS 2017:防御对抗性攻击](https://www.kaggle.com/c/nips-2017-defense-against-adversarial-attack/data)
* [NIPS 2018:对抗性视觉挑战赛](https://www.crowdai.org/challenges)
* [GeekPwn CAAD 2018](http://2018.geekpwn.org/en/index.html#4).
* [IJCAI-19 阿里巴巴对抗性 AI 挑战赛](https://tianchi.aliyun.com/markets/tianchi/ijcai19_en)
* [GeekPwn CAAD 2019](http://www.geekpwn.org/zh/index.html)
* [Positive Hack Days 2019:AI CTF](https://2019.phdays.com/en/program/contests/aI-ctf/)
* [Positive Hack Days 2021:AI CTF](https://2021.phdays.com/en/program/contests/ai-track/)
* [Positive Hack Days 2022:AI CTF](https://ai.ctf.su/)
* [UTCTF 2019(FaceSafe、Bot Protection IV 任务)](https://github.com/utisss/UTCTF-19)
* [vishwaCTF21(Good Driver Bad Driver 任务)](https://vishwactf.com/)
* [AI/LLM 利用挑战赛(AI CTF 实验室)](https://academy.8ksec.io/course/ai-exploitation-challenges)
## 框架
* [**adversarial-robustness-toolbox**](https://github.com/IBM/adversarial-robustness-toolbox)
* [**foolbox**](https://github.com/bethgelab/foolbox)
* [**cleverhans**](https://github.com/tensorflow/cleverhans)
标签:AI安全, Chat Copilot, 人工智能, 内核模块, 凭据扫描, 威胁建模, 安全资源, 密钥泄露防护, 机器学习安全, 用户模式Hook绕过, 逆向工具