Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced Data
Utkarsh Ojha, Krishna Kumar Singh, Cho-Jui Hsieh, Yong Jae Lee
Abstract
We propose a novel unsupervised generative model that learns to disentangle object identity from other low-level aspects in class-imbalanced data. We first investigate the issues surrounding the assumptions about uniformity made by InfoGAN, and demonstrate its ineffectiveness to properly disentangle object identity in imbalanced data. Our key idea is to make the discovery of the discrete latent factor of variation invariant to identity-preserving transformations in real images, and use that as a signal to learn the appropriate latent distribution representing object identity. Experiments on both artificial (MNIST, 3D cars, 3D chairs, ShapeNet) and real-world (YouTube-Faces) imbalanced datasets demonstrate the effectiveness of our method in disentangling object identity as a latent factor of variation.
BibTeX
@inproceedings{NEURIPS2020_d1e39c9b,
author = {Ojha, Utkarsh and Singh, Krishna Kumar and Hsieh, Cho-Jui and Lee, Yong Jae},
booktitle = {Advances in Neural Information Processing Systems},
editor = {H. Larochelle and M. Ranzato and R. Hadsell and M.F. Balcan and H. Lin},
pages = {18063--18075},
publisher = {Curran Associates, Inc.},
title = {Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced Data},
url = {https://proceedings.neurips.cc/paper_files/paper/2020/file/d1e39c9bda5c80ac3d8ea9d658163967-Paper.pdf},
volume = {33},
year = {2020}
}