DeepXplore: Automated Whitebox Testing of Deep Learning Systems · HackerPapers discussions

DeepXplore: Automated Whitebox Testing of Deep Learning Systems — mentioned by stochastic_monk

Sat, 10 Mar 2018 12:58:45 GMT

I think that DeepXplore [0] is on the right track. I think that perhaps some methodological improvements could likely be made, but applying an adversarial, security-like approach to deep neural networks is a way to build fault-tolerance into these methods. I'll keep hoping that things like capsules and "smarter" network design/training and data augmentation will eventually, de novo, help add "safe" generalizability. Perhaps optimizing accuracy/minimizing loss in a more broadly, uniformly random way rather than in optimizing accuracy, weighted by frequency of observations. [0]…

Paper and discussion history. A mention is not an endorsement.

DeepXplore: Automated Whitebox Testing of Deep Learning Systems — mentioned by wpasc

Mon, 26 Feb 2018 19:50:20 GMT

I read a great, great paper about training systems against this. Obfuscation is currently easy against AI, but obfuscation that fools an AI and a human could still detect (Purposeful perturbations). This can be pre-trained against by incorporating such perturbations programmatically during training. Link: https://arxiv.org/pdf/1705.06640.pdf

Paper and discussion history. A mention is not an endorsement.

DeepXplore: Automated Whitebox Testing of Deep Learning Systems — HN submission

Tue, 06 Jun 2017 12:45:23 GMT

DeepXplore: Automated Whitebox Testing of Deep Learning Systems

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.