Robustness of classifiers_from_adversarial_to_random_noise
2 likes789 views
The document discusses the robustness of classifiers in the context of deep learning, specifically addressing their performance against adversarial and random noise inputs. It examines the state-of-the-art findings and various empirical examples in the domain. The paper highlights the implications for improving classifier resilience in machine learning applications.
![(1)
• [1, 10, 19]
• DL adversarial example [17]
• DL adversarial example
• [3, 5, 14, 18]
• robust network [6, 8, 20, 13, 15, 12]
8](https://image.slidesharecdn.com/robustnessofclassifiersfromadversarialtorandomnoise-170219125751/85/Robustness-of-classifiers_from_adversarial_to_random_noise-8-320.jpg)
![(2)
• [18] empirical adversarial example
pixel
• [3] random adversarial
9](https://image.slidesharecdn.com/robustnessofclassifiersfromadversarialtorandomnoise-170219125751/85/Robustness-of-classifiers_from_adversarial_to_random_noise-9-320.jpg)
![1: 2-1 (2)
• [13]
• 1000
35](https://image.slidesharecdn.com/robustnessofclassifiersfromadversarialtorandomnoise-170219125751/85/Robustness-of-classifiers_from_adversarial_to_random_noise-35-320.jpg)
Robustness of classifiers_from_adversarial_to_random_noise
- 1. Robustness of classifiers: from adversarial to random noise hskksk @ 2017/2/3 1
- 3. 3
- 4. Fawzi, A., Moosavi-Dezfooli, S.-M., & Frossard, P. (2016). Robustness of classifiers: from adversarial to random noise. In NIPS (pp. 1624–1632). 4
- 5. • Deep learning state-of-the-art • Adversarial ( ) • • Adversarial 5
- 8. (1) • [1, 10, 19] • DL adversarial example [17] • DL adversarial example • [3, 5, 14, 18] • robust network [6, 8, 20, 13, 15, 12] 8
- 9. (2) • [18] empirical adversarial example pixel • [3] random adversarial 9
- 10. • semi-random • semi-random random/worst-case • worst-case semi-random • 10
- 11. nota%on 11
- 13. • • • 13
- 15. 15
- 16. 16
- 17. • • 17
- 18. 1: • 18
- 19. • 1 • • • 19
- 21. (m=1 ) • 21
- 22. 1 • • m 22
- 23. 23
- 25. ( ) • • 25
- 26. (1) • bound • • i j worst-case 26
- 27. (2) • worst-case radius: • : • • 27
- 28. 2: ( ) • • 28
- 30. • • ( ) • 30
- 32. 2-1: (2) • • global 2 • 2 ( ) 32
- 33. 33
- 34. 1: 2-1 (1) • • • 1 1 34
- 35. 1: 2-1 (2) • [13] • 1000 35
- 36. 1 (1) • 1 2-1 36
- 37. 1 (2) • 37
- 38. 2: (5) • • • 38
- 39. 2 (1) 39
- 43. • → adversarial • robust • state-of-the-art semi-random 43