A Revisit of Total Correlation in Disentangled Variational Auto-Encoder with Partial Disentanglement
This is an incremental improvement for researchers in representation learning, addressing flexibility in disentanglement for complex datasets.
The authors tackled the problem that enforcing full independence in disentangled variational auto-encoders (VAEs) can be too strict for datasets with entangled factors, by developing the Partially Disentangled VAE (PDisVAE) that generalizes total correlation to partial correlation for group-wise independence, and validation on synthetic and real-world datasets showed it discovers valuable information not found with fully disentangled VAEs.
A fully disentangled variational auto-encoder (VAE) aims to identify disentangled latent components from observations. However, enforcing full independence between all latent components may be too strict for certain datasets. In some cases, multiple factors may be entangled together in a non-separable manner, or a single independent semantic meaning could be represented by multiple latent components within a higher-dimensional manifold. To address such scenarios with greater flexibility, we develop the Partially Disentangled VAE (PDisVAE), which generalizes the total correlation (TC) term in fully disentangled VAEs to a partial correlation (PC) term. This framework can handle group-wise independence and can naturally reduce to either the standard VAE or the fully disentangled VAE. Validation through three synthetic experiments demonstrates the correctness and practicality of PDisVAE. When applied to real-world datasets, PDisVAE discovers valuable information that is difficult to find using fully disentangled VAEs, implying its versatility and effectiveness.