[paper]The Limitations of Deep Learning in Adversarial Settings(JSMA）
與之前的基於提高原始類別標記的損失函數或者降低目標類別標籤概率的損失函數的方式不同，這篇文章提出利用輸入特徵到輸出值之間的對抗顯著性，達到只需修改少量的輸入值即可誤分類的目的。 換句話說，之前的對抗樣本的擾動方向都是損失函數的梯度方向（無論是原始類別標記的損失函數還是目標類別標記的損失函數），該論文生成的對抗樣本的擾動方向是目標類別標記的預測值的梯度方向，作者將這個梯度稱爲前向梯度（forward
相關文章
- 1. The Limitations of Deep Learning in Adversarial Settings
- 2. 關於The Limitations of Deep Learning in Adversarial Settings的理解
- 3. 【論文回顧】The Limitations of Deep Learning in Adversarial Settings
- 4. The Limitations of Deep Learning in Adversarial Settings論文筆記
- 5. 對The Limitations of Deep Learning in Adversarial Settings理解
- 6. Application of deep learning in Industrial area
- 7. [論文解讀]Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
- 8. [轉載][paper]Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
- 9. Exploring the teaching of deep learning in neural networks
- 10. Evaluation of Deep Learning Toolkits
- 更多相關文章...
- • SQL IN 操作符 - SQL 教程
- • Swift for-in 循環 - Swift 教程
- • Java Agent入門實戰（一）-Instrumentation介紹與使用
- • Java Agent入門實戰（三）-JVM Attach原理與使用
相關標籤/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。
歡迎關注本站公眾號,獲取更多信息
相關文章
- 1. The Limitations of Deep Learning in Adversarial Settings
- 2. 關於The Limitations of Deep Learning in Adversarial Settings的理解
- 3. 【論文回顧】The Limitations of Deep Learning in Adversarial Settings
- 4. The Limitations of Deep Learning in Adversarial Settings論文筆記
- 5. 對The Limitations of Deep Learning in Adversarial Settings理解
- 6. Application of deep learning in Industrial area
- 7. [論文解讀]Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
- 8. [轉載][paper]Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
- 9. Exploring the teaching of deep learning in neural networks
- 10. Evaluation of Deep Learning Toolkits