Concept
Reparameterization trick
The reparameterization trick is a technique for making a random node in a computation graph differentiable with respect to the parameters that define its distribution. The problem it solves: if you sample z ~ N(mu, sigma^2), you cannot backpropagate through the sampling…
The rest of “Reparameterization trick” is a premium feature: every concept in the library gets a precise, practitioner-focused write-up like this one, cross-linked straight from the paper summaries that use it.
Log in to unlock→