Anonymous ID: e22891 Sept. 18, 2018, 2:18 p.m. No.3076891   🗄️.is 🔗kun   >>6934 >>7007 >>7071 >>7289 >>7354

>>3076664

>>>/patriotsfight/259

 

Image recognition is woefully bad at handling adversarial input. It'd be trivial to create false negatives that Facebook's "AI" would miss.

 

https://cacm.acm.org/magazines/2018/7/229030-making-machine-learning-robust-against-adversarial-inputs/fulltext

https://archive.fo/EwIfF

 

>Adversarial Attacks on Machine Learning Models

http://nesl.ee.ucla.edu/uploads/document/slidesupload/537/afm_adversarial_tutorial_workshop_external.pdf

https://archive.fo/tuDZ1

 

Pics related.

 

"NIPS 2017: Targeted Adversarial Attack Develop an adversarial attack that causes image classifiers to predict a specific target class"

https://www.kaggle.com/c/nips-2017-targeted-adversarial-attack

https://archive.fo/I8zaq