y0ast/Variational-Autoencoder

reconstructed_x

nationalflag opened this issue · 1 comments

reconstructed_x seems just the mean of the decode image, don't we need to combine it with the variance to obtain the decode image?

y0ast commented

When the input consists of continuous variables, we indeed try to find the mean and variance of each feature for the output. However when the input is binary (as in the case of MNIST) a single mean fully parametrizes it (binomial distribution).