[paper]AdvJND:Generating Adversarial Examples with Just Noticeable Difference
生成对抗样本有两个要求：攻击成功率和图像保真度指标。 增加扰动可以确保对抗样本的攻击成功率很高； 但是生成的对抗样本隐蔽性很差。 为了在攻击成功率和图像保真度之间取折衷，提出了一种名为AdvJND的方法，该方法在生成对抗样本时在失真函数的约束下添加了视觉模型系数，该系数用来衡量视觉上的差异。AdvJND算法生成的对抗样本产生的梯度分布与原始输入相似。该方法可以认为是一种辅助生成方法，用来改善生成算
相关文章
- 1. [advGAN]Generating Adversarial Examples With Adversarial Networks
- 2. Generating Adversarial Examples with Adversarial Networks
- 3. Generating Adversarial Examples with Adversarial Networks阅读笔记
- 4. Paper Review: Adversarial Examples
- 5. Certified Robustness to Adversarial Examples with Differential Privacy
- 6. SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES
- 7. Explaining and Harnessing Adversarial Examples
- 8. [paper]SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES
- 9. Adversarial Examples 对抗样本
- 10. CS231n Lecture 16 | Adversarial Examples and Adversarial Training
- 更多相关文章...
- • XPath Examples - XPath 教程
- • XSLT
- • RxJava操作符（一）Creating Observables
- • RxJava操作符（七）Conditional and Boolean
相关标签/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
欢迎关注本站公众号,获取更多信息
相关文章
- 1. [advGAN]Generating Adversarial Examples With Adversarial Networks
- 2. Generating Adversarial Examples with Adversarial Networks
- 3. Generating Adversarial Examples with Adversarial Networks阅读笔记
- 4. Paper Review: Adversarial Examples
- 5. Certified Robustness to Adversarial Examples with Differential Privacy
- 6. SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES
- 7. Explaining and Harnessing Adversarial Examples
- 8. [paper]SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES
- 9. Adversarial Examples 对抗样本
- 10. CS231n Lecture 16 | Adversarial Examples and Adversarial Training