Mode collapse on MNIST #9
Description
In the MNIST experiment, I found that the generated numbers in each category were the same, with no diversity. I've checked the init_noise, and it is random. Is there something wrong with the code or the algorithm's limitation?