Reparameterization gradient Created 2026-10-06 Updated 2026-10-25
A reparameterization gradient differentiates an expected value by representing the simulated random variable as a differentiable transformation of parameter-independent noise. For , the identity requires justified differentiation under the integral sign. For a normal distribution, use with drawn from the standard normal distribution.
For a parameter-independent integrand , the displayed score-function gradient estimator follows by differentiating the sampling probability density function or probability mass function, using . It requires conditions permitting differentiation under the integral sign, or the corresponding sum, and control of support changes. It applies to discrete choices without pretending that their sampled values are differentiable paths. If , add .
Scorer Hi function 2026-10-05
For real , the integral representation is
Using differentiation under the integral sign twice and then integration by parts shows . The cubic decay makes the integral converge for every real .