I can not remember the name of the paper, but researchers have recently discovered that neural networks can fairly easily be disrupted by carefully perturbing t

I can not remember the name of the paper, but researchers have recently discovered that neural networks can fairly easily be disrupted by carefully perturbing the pixels in incredibly slight amounts. Completely invisible to a human, but can ruin the ability of neural networks to classify it.