Good video. Just a quick question: When we take gradients of the ELBO wrt the variational parameters, can we actually write the MC Approximation given in 10:22 ? gradients of expectation will not be equal to the Expectation of gradients when we're sampling from the same distribution wrt which we're taking gradients. In ADVI, we end up reparametrizing to Standard Normal DIstribution. Any explanation would be helpful
Good video. Just a quick question: When we take gradients of the ELBO wrt the variational parameters, can we actually write the MC Approximation given in 10:22 ? gradients of expectation will not be equal to the Expectation of gradients when we're sampling from the same distribution wrt which we're taking gradients. In ADVI, we end up reparametrizing to Standard Normal DIstribution. Any explanation would be helpful