CLEVER score

The CLEVER (Cross Lipschitz Extreme Value for nEtwork Robustness) score is a way of measuring the robustness of an artificial neural network towards adversarial attacks.^[1] It was developed by a team at the MIT-IBM Watson AI Lab in IBM Research and first presented at the 2018 International Conference on Learning Representations.^[2] It was mentioned and reviewed by Ian Goodfellow^[3] as well. It was adopted into a educational game Fool The Bank^[4] by Narendra Nath Joshi,^[5] Abhishek Bhandwaldar and Casey Dugan

References

↑ Weng, Tsui-Wei (2018). "Evaluating the robustness of neural networks: An extreme value theory approach". arXiv:1801.10578 [stat.ML].
↑ "A CLEVER Way to Resist Adversarial Attack". May 2, 2018. https://www.ibm.com/blogs/research/2018/05/clever-adversarial-attack/.
↑ "Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach". 10 February 2022. https://openreview.net/forum?id=BkUHlMZ0b.
↑ "Fool the Bank - IBM Research". https://foolthebank.mybluemix.net/.
↑ "Narendra Nath Joshi". http://nnjoshi.co.

0.00

(0 votes)

Original source: https://en.wikipedia.org/wiki/CLEVER score. Read more

[1] Weng, Tsui-Wei (2018). "Evaluating the robustness of neural networks: An extreme value theory approach". arXiv:1801.10578 [stat.ML].

[2] "A CLEVER Way to Resist Adversarial Attack". May 2, 2018. https://www.ibm.com/blogs/research/2018/05/clever-adversarial-attack/.

[3] "Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach". 10 February 2022. https://openreview.net/forum?id=BkUHlMZ0b.

[4] "Fool the Bank - IBM Research". https://foolthebank.mybluemix.net/.

[5] "Narendra Nath Joshi". http://nnjoshi.co.

[1]

[2]

[3]

[4]

[5]