jmyoon1

I'm curious about this, too. Have you found any reason? logvar does simply seems to reduce over training steps if trained. I see that this affects to the decoder variance,...