Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
This work addresses the problem of learning interpretable and transferable representations for machine learning applications, though it appears incremental as it builds on existing variational methods.
The paper tackles unsupervised learning of disentangled representations from unlabeled observations by proposing a variational inference approach with a new regularizer and metric, resulting in significant improvements in disentanglement and reconstruction quality over existing methods.
Disentangled representations, where the higher level data generative factors are reflected in disjoint latent dimensions, offer several benefits such as ease of deriving invariant representations, transferability to other tasks, interpretability, etc. We consider the problem of unsupervised learning of disentangled representations from large pool of unlabeled observations, and propose a variational inference based approach to infer disentangled latent factors. We introduce a regularizer on the expectation of the approximate posterior over observed data that encourages the disentanglement. We also propose a new disentanglement metric which is better aligned with the qualitative disentanglement observed in the decoder's output. We empirically observe significant improvement over existing methods in terms of both disentanglement and data likelihood (reconstruction quality).