5月第三周
- Machine learning has a backdoor problem https://bdtechtalks.com/2022/05/23/machine-learning-undetectable-backdoors/
- (De)ToxiGen: Leveraging large language models to build more robust hate speech detection tools https://www.microsoft.com/en-us/research/blog/detoxigen-leveraging-large-language-models-to-build-more-robust-hate-speech-detection-tools/?OCID=msr_blog_DeToxiGen_TW
- Introducing two new datasets to help measure fairness and mitigate AI bias https://ai.facebook.com/blog/measure-fairness-and-mitigate-ai-bias/
- Learning Machine Learning Part 3: Attacking Black Box Models https://posts.specterops.io/learning-machine-learning-part-3-attacking-black-box-models-3efffc256909
往期回顾：
2021年10月到12月
1.【论文】关于图神经网络 (Graph Neural Network) 生成的图嵌入 (Graph Embedding) 中的 隐私泄漏的文章Inference Attacks Against Graph Neural Networks
目前该工作已被 Usenix Security 2022 录用 本文着重分析了图嵌入共享过程中可能存在的隐私泄露问题 作者提出了三种攻击场景，即图属性推断攻击、子图推断攻击，以及图重构攻击，并系统地评估了这三种攻击的有效性 最后，作者提出了基于图嵌入扰动的方法来降低这些攻击带来的风险
https://arxiv.org/pdf/2110.02631.pdf
2.AAAI’22|基于验证外源特征的模型偷盗防御 https://zhuanlan.zhihu.com/p/427111937
3.SeqAttack: Token分类模型对抗攻击框架 https://github.com/WalterSimoncini/SeqAttack
4.腾讯朱雀实验室推出代码防护技术Deep Puzzling，让代码更难被猜透 https://www.jiqizhixin.com/articles/2021-11-29
5.机器学习DGA域名检测 https://xz.aliyun.com/t/10522
- 没有银弹-AI安全领域的安全与隐私 https://www.anquanke.com/post/id/258704
- AI生成代码对网络安全的影响 https://www.cfr.org/blog/ai-code-generation-and-its-implications-cybersecurity
- AI赋能windows恶意软件检测 https://www.anquanke.com/post/id/256969