storchastic
storchastic copied to clipboard
Implement Quadratic Approximation baseline for reparameterization
See https://arxiv.org/pdf/2007.14634.pdf .
Implementation is a bit interesting as it requires adding a loss, while reparameterization doesn't usually do this. So this just uses an additive loss on that node (zero mean, as it is a control variate!), but should work fine regardless.