DynamicVAE: Decoupling Reconstruction Error and Disentangled Representation Learning

TL;DR

DynamicVAE introduces a novel approach that dynamically adjusts the KL-divergence weight during training to decouple disentanglement from reconstruction accuracy, leading to improved performance in both aspects.

Contribution

The paper proposes DynamicVAE, which uses a stage-wise, adaptive control of the KL-divergence weight to separate disentanglement from reconstruction, overcoming limitations of existing methods.

Findings

Significantly improves reconstruction accuracy

Achieves comparable disentanglement to state-of-the-art methods

Successfully decouples disentanglement from reconstruction trade-off

Abstract

This paper challenges the common assumption that the weight $\beta$, in $\beta$-VAE, should be larger than $1$ in order to effectively disentangle latent factors. We demonstrate that $\beta$-VAE, with $\beta < 1$, can not only attain good disentanglement but also significantly improve reconstruction accuracy via dynamic control. The paper removes the inherent trade-off between reconstruction accuracy and disentanglement for $\beta$-VAE. Existing methods, such as $\beta$-VAE and FactorVAE, assign a large weight to the KL-divergence term in the objective function, leading to high reconstruction errors for the sake of better disentanglement. To mitigate this problem, a ControlVAE has recently been developed that dynamically tunes the KL-divergence weight in an attempt to control the trade-off to more a favorable point. However, ControlVAE fails to eliminate the conflict between the need for a large $\beta$ (for disentanglement) and the need for a small $\beta$. Instead, we propose DynamicVAE that maintains a different $\beta$ at different stages of training, thereby decoupling disentanglement and reconstruction accuracy. In order to evolve the weight, $\beta$, along a trajectory that enables such decoupling, DynamicVAE leverages a modified incremental PI (proportional-integral) controller, and employs a moving average as well as a hybrid annealing method to evolve the value of KL-divergence smoothly in a tightly controlled fashion. We theoretically prove the stability of the proposed approach. Evaluation results on three benchmark datasets demonstrate that DynamicVAE significantly improves the reconstruction accuracy while achieving disentanglement comparable to the best of existing methods. The results verify that our method can separate disentangled representation learning and reconstruction, removing the inherent tension between the two.