SSVae tutorial. Which is the classifier network?


#1

Hi, Im reading the SSVAE tutorial. Does the encoder_y or the decoder_z correspond to a neural digit classifier?
http://pyro.ai/examples/ss-vae.html
In the tutorial it states that:

We structure the components of our Guide qϕ(.) as follows:

  • qϕ(y | x)=Cat(y | αϕ(x)): parameterized multinomial (or categorical) distribution; αϕ(x)corresponds to encoder_y in the code

  • qϕ(z | x,y)=N(z | μϕ(x,y),σ2ϕ(x,y)): parameterized normal distribution; μϕ(x,y) and σ2ϕ(x,y) correspond to the neural digit classifier encoder_z in the code

But it seems to me that the encoder_y is the classifier. Am i wrong?


#2

you’re correct; encoder_y is the digit classifier, encoder_z approximates the latent distribution (‘style’ classifier). thanks for catching that