I am trying to implement the boosting of variational inference with a neural network. For regular variational inference, I could implement it with pyro’s hidden layer (pyro.contrib.bnn.HiddenLayer). But in boosting as I will have multiple mean values how should I change the hidden layer class in pyro. Or is there another way?
The logical solution that I think is to customize the Hiddenlayer Class that I imported “import pyro.contrib.bnn as bnn”. Is this correct or are there any other ways to solve this?