One post tagged with "CNF"

Introduction​

Generative models aim to transform a simple base distribution, such as a Gaussian, into a complex target data distribution. The core idea is to construct a generator that maps samples from the base distribution into data space. To compute likelihoods, this generator must be invertible. If a generator transforms $Z$ into $X = G(Z)$, then $Z = G^{-1}(X)$ must exist, and the density is computed using the change of variables:

$P_X(X) = P_Z(G^{-1}(X)) \left|\det(\partial G^{-1}(X)/\partial X)\right|$