Independent Subspace Analysis for Unsupervised Learning of Disentangled Representations
This work addresses the challenge of learning interpretable and structured representations in unsupervised machine learning, which is incremental as it builds on the VAE framework by modifying the probabilistic model rather than the cost function.
The paper tackles the problem of unsupervised learning of disentangled representations in Variational Autoencoders (VAEs) by showing that existing methods like beta-VAE cause over-pruning and over-orthogonalization, and proposes a structured latent prior that resolves unidentifiability, encourages disentanglement, and significantly mitigates the trade-off between reconstruction loss and disentanglement over state-of-the-art methods.
Recently there has been an increased interest in unsupervised learning of disentangled representations using the Variational Autoencoder (VAE) framework. Most of the existing work has focused largely on modifying the variational cost function to achieve this goal. We first show that these modifications, e.g. beta-VAE, simplify the tendency of variational inference to underfit causing pathological over-pruning and over-orthogonalization of learned components. Second we propose a complementary approach: to modify the probabilistic model with a structured latent prior. This prior allows to discover latent variable representations that are structured into a hierarchy of independent vector spaces. The proposed prior has three major advantages: First, in contrast to the standard VAE normal prior the proposed prior is not rotationally invariant. This resolves the problem of unidentifiability of the standard VAE normal prior. Second, we demonstrate that the proposed prior encourages a disentangled latent representation which facilitates learning of disentangled representations. Third, extensive quantitative experiments demonstrate that the prior significantly mitigates the trade-off between reconstruction loss and disentanglement over the state of the art.