Variational Autoencoders with Decremental Information Bottleneck for Disentanglement

Jiantao Wu, Shentong Mo, Muhammad Awais, Sara Atito, xingshen zhang, Lin Wang, Xiang Yang

Keywords: disentanglement, variational autoencoders

TL;DR: We present a novel decremental variational autoencoder with disentanglement-invariant transformations, termed DeVAE, for balancing disentanglement and reconstruction fidelity.

Abstract: One major challenge of disentanglement learning with variational autoencoders is the trade-off between disentanglement and reconstruction fidelity. Previous methods, spreading the conflict of disentanglement and reconstruction in time, will lose the constraint of disentanglement when expanding the information bottleneck, which causes the information diffusion problem. To tackle this issue, we present a novel decremental variational autoencoder with disentanglement-invariant transformations to spread the conflict on multiple latent spaces, termed DeVAE, for balancing disentanglement and reconstruction fidelity by decreasing the information bottleneck of diverse latent spaces gradually. Benefiting from the multiple latent spaces, DeVAE allows simultaneous optimization of multiple objectives to optimize reconstruction while keeping the constraint of disentanglement, avoiding information diffusion. DeVAE is also compatible with large models with high-dimension latent space. Experimental results on dSprites and Shapes3D that DeVAE achieves the best performance on both disentanglement and reconstruction.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning