Training MLP on 2D-binary classification.
$ python main.pyThis script creates following GIF.
Decision boundary and entropy plots on 2D-binary classification by MLPs trained with Adam and VOGN.
VOGN optimizes the posterior distribution of each weight (i.e., mean and variance of the Gaussian). A model with the mean weights draws the red boundary, and models with the MC samples from the posterior distribution draw light red boundaries. VOGN converges to a similar solution as Adam while keeping uncertainty in its predictions.