When you buy furniture from IKEA, you really buy a representation of the furniture you want. You can regenerate a close approximation of the furniture from the representation you purchase, but never the one that was before it was encoded.
@nolovedeeplearning @emtiyaz @mathieualain @guy
Iām gonna slightly shamelessly self-plug one of our recent works here too, to share my recently found enthusiasm about #ProbabilisticCircuit :)!
Intuitively, we blur the boundaries again & investigate what PCs with differentiable sampling can do, in an effort to compare with #autoencoding, #GANs etc that made generative NNs more prominent.
Paper is available here: https://proceedings.mlr.press/v181/lang22a/lang22a.pdf
With @sbraun & @kerstingAIML
#ProbabilisticCircuit #autoencoding #GANs