Identifying through Flows for Recovering Latent Representations

Shen Li, Bryan Hooi, Gim Hee Lee

Identifiability, or recovery of the true latent representations from which the observed data originates, is de facto a fundamental goal of representation learning. Yet, most deep g…