Uncertainty estimation on Mnist dataset
This is a PyTorch implementation of Dropout Uncertainty on Mnist. The experiment setting is based on Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning at 5.2 Model Uncertainty in Classification Tasks.
- Install Pytorch Pytorch
conda install pytorch torchvision -c pytorch
- Clone this repository
git clone https://github.com/andyhahaha/Uncertainty_Mnist
- Train Lenet standard and Lenet dropout
python main.py --mode 0
- Test Lenet standard and Lenet dropout
python main.py --mode 1
- Test the Lenet dropout on rotated Mnist image
python main.py --mode 2
These results show the uncertainty of different rotated digits.
| 0 | 1 |
 |
 |
| 2 | 3 |
 |
 |
| 4 | 5 |
 |
 |
| 6 | 7 |
 |
 |
| 8 | 9 |
 |
 |