# Decision-based adversarial attacks: Reliable attacks against black-box machine learning models
###### tags: `paper` `attack` `evasion attack` `boundary attack` `adversarial`

## source
- [[论文总结] Boundary Attack](https://zhuanlan.zhihu.com/p/377633699)
- [最大熵模型](https://luweikxy.gitbook.io/machine-learning-notes/linear-model/maximum-entropy-model#%E7%9B%B4%E8%A7%82%E7%90%86%E8%A7%A3%E6%9C%80%E5%A4%A7%E7%86%B5)
- [近年攻防method總匯](https://zhuanlan.zhihu.com/p/135374750)

- [Why Do Adversarial Attacks Transfer? Explaining Transferability of Evasion and Poisoning Attacks](https://www.usenix.org/conference/usenixsecurity19/presentation/demontis)
- [中文分享](https://blog.csdn.net/qq_26130991/article/details/109516099)

- [NIPS 2018 untargeted 冠軍 github](https://github.com/luizgh/avc_nips_2018)

- [Delving into transferable adversarial examples and black-box attacks](https://blog.csdn.net/qq_35414569/article/details/82383788?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522162615837816780271525811%2522%252C%2522scm%2522%253A%252220140713.130102334..%2522%257D&request_id=162615837816780271525811&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~baidu_landing_v2~default-1-82383788.pc_search_result_before_js&utm_term=%E6%94%BB%E5%87%BB%E8%BF%81%E7%A7%BB%E6%80%A7)

- [Decoupling Direction and Norm for Efficient Gradient-Based L2 Adversarial Attacks and Defenses](https://arxiv.org/abs/1811.09600)
- 改良此篇論文之boundary attack， 奪得 [NIPS 2018]() untargeted 組冠軍
- [hackmd 筆記](https://hackmd.io/719H2FT3RFu9N_Jh7J_F4Q)
- [An Efficient Pre-processing Method to Eliminate Adversarial Effects](https://arxiv.org/pdf/1905.08614.pdf)
- 對圖片先做WebP壓縮再左右翻轉
- 對抗樣本有可移植性
- https://zhuanlan.zhihu.com/p/217683614