Stochastic Backpropagation through Mixture Density Distributions
arXiv: Neural and Evolutionary Computing, Volume abs/1607.05690, 2016.
The ability to backpropagate stochastic gradients through continuous latent distributions has been crucial to the emergence of variational autoencoders and stochastic gradient variational Bayes. The key ingredient is an unbiased and low-variance way of estimating gradients with respect to distribution parameters from gradients evaluated a...More
PPT (Upload PPT)