Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced Data
This addresses the challenge of learning disentangled representations for imbalanced data, which is common in real-world applications, though it is an incremental improvement over InfoGAN.
The paper tackles the problem of unsupervised disentangled representation learning in class-imbalanced data, showing that InfoGAN fails to properly disentangle object identity in such settings. It proposes Elastic-InfoGAN, which learns to make the discovery of discrete latent factors invariant to identity-preserving transformations, achieving effective disentanglement on datasets like MNIST and YouTube-Faces.
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.