PyTorch-BayesianCNN
PyTorch-BayesianCNN copied to clipboard
What is the function of the fixed prior value (i.e., prior_mu and prior_sigma) ?
Can the prior distribution be arbitrarily set? In the CNN, the weight can be optimized by the back propagate. However, in the BCNN, the W_mu and W_sigma always seem to be approximated the initial prior distribution?