A Spectral Regularizer for Unsupervised Disentanglement

12/04/2018
by   Aditya Ramesh, et al.
10

Generative models that learn to associate variations in the output along isolated attributes with coordinate directions in latent space are said to possess disentangled representations. This has been a recent topic of great interest, but remains poorly understood. We show that even for GANs that do not possess disentangled representations, one can find paths in latent space over which local disentanglement occurs. These paths are determined by the leading right-singular vectors of the Jacobian of the generator with respect to its input. We describe an efficient regularizer that aligns these vectors with the coordinate axes, and show that it can be used to induce high-quality, disentangled representations in GANs, in a completely unsupervised manner.

READ FULL TEXT

page 5

page 6

page 7

page 13

page 14

page 22

research
02/21/2021

Do Generative Models Know Disentanglement? Contrastive Learning is All You Need

Disentangled generative models are typically trained with an extra regul...
research
05/15/2020

Face Identity Disentanglement via Latent Space Mapping

Learning disentangled representations of data is a fundamental problem i...
research
04/07/2021

Where and What? Examining Interpretable Disentangled Representations

Capturing interpretable variations has long been one of the goals in dis...
research
02/13/2022

Unsupervised Disentanglement with Tensor Product Representations on the Torus

The current methods for learning representations with auto-encoders almo...
research
11/12/2018

Improving Generalization for Abstract Reasoning Tasks Using Disentangled Feature Representations

In this work we explore the generalization characteristics of unsupervis...
research
09/27/2021

WarpedGANSpace: Finding non-linear RBF paths in GAN latent space

This work addresses the problem of discovering, in an unsupervised manne...
research
05/26/2022

Analyzing the Latent Space of GAN through Local Dimension Estimation

The impressive success of style-based GANs (StyleGANs) in high-fidelity ...

Please sign up or login with your details

Forgot password? Click here to reset