GlassyWing / Uncertainty-Quantification-Result

Uncertainty quantification experiment results of classification tasks.

Uncertainty Quantification

Uncertainty quantification experiment results of classification tasks. "What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?", NIPS 2017

Mnist experiment

The normal images:

The corrupted images:

The more corrupt the image, the higher the cognitive uncertainty. It seems that this is a very simple and effective method.