flâneur

RescaledKL.tm

danilorezende.com · 1,490 words · saved by 1 readers

N/A

Short Notes on Variational Bounds with Rescaled Terms 1 1 Short Notes on Variational Bounds with Rescaled Terms R Learning a generative model of the form p(x) = dzp(x; z) for x 2 Rd and z 2 Rm involves P maximizing the likelihood of the data L = x2Data ln p(x). P When using variational approximations, we instead optimize the variational bound F = x2Data F(x), where F(x) =¡E q(z jx)[ln p(xjz)] + KL(q(z jx)kp(z)) > ¡ln p(x);…

saved by

related reading