This repository reimplemented "MC Dropout" by tensorflow 2.0 Eager Extension.
paper: Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
(https://arxiv.org/pdf/1506.02142.pdf)
Here are the demo results. The condition is below.
Dataset : CIFAR-10
Model Training :
Epoch : 10
Dropout Rate : 0.5
Prediction :
Sample Number : 100
The higher entropy, the more difficult it is to predict.
The top 10 for the case of "airplane" and "bird" is shown.
・High entropy top 10
・Low entropy top 10