An issue in Neural Networks, wherein a model does not know when its wrong. A cost function is trained to be confident in specific outputs, there is no option to be unsure of an answer.

Deep Neural Rejection is an attempt to fix this.