Learning Disentangled Representations with the Wasserstein Autoencoder

Benoit Gaujac, Ilya Feige, David Barber

Abstract: Disentangled representation learning has undoubtedly benefited from objective function surgery. However, a delicate balancing act of tuning is still required in order to trade off reconstruction fidelity versus disentanglement. Building on previous successes of penalizing the total correlation in the latent variables, we propose TCWAE (Total Correlation Wasserstein Autoencoder). Working in the WAE paradigm naturally enables the separation of the total-correlation term, thus providing disentanglement control over the learned representation, while offering more flexibility in the choice of reconstruction cost. We propose two variants using different KL estimators and analyse in turn the impact of having different ground cost functions and latent regularization terms. Extensive quantitative comparisons on data sets with known generative factors shows that our methods present competitive results relative to state-of-the-art techniques. We further study the trade off between disentanglement and reconstruction on more-difficult data sets with unknown generative factors, where the flexibility of the WAE paradigm leads to improved reconstructions.