TY - GEN
T1 - Geometry of Deep Generative Models for Disentangled Representations
AU - Shukla, Ankita
AU - Uppal, Shagun
AU - Bhagat, Sarthak
AU - Anand, Saket
AU - Turaga, Pavan
N1 - Funding Information:
This work is supported in part by Infosys Center for Artificial intelligence at IIIT Delhi, India.
Publisher Copyright:
© 2018 ACM.
PY - 2018/12/18
Y1 - 2018/12/18
N2 - Deep generative models like variational autoencoders approximate the intrinsic geometry of high dimensional data manifolds by learning a set of low-dimensional latent-space variables and an embedding function. The geometrical properties of these latent spaces has been studied under the lens of Riemannian geometry; via analysis of the non-linearity of the generator function. In new developments, deep generative models have been used for learning semantically meaningful 'disentangled' representations; that capture task relevant attributes while being invariant to other attributes. In this work, we explore the geometry of popular generative models for disentangled representation learning. We use several metrics to compare the properties of latent spaces of disentangled representation models in terms of class separability and curvature of the latent-space. The proposed study establishes that the class distinguishable features in the disentangled latent space exhibits higher curvature as opposed to a variational autoencoder. We evaluate and compare the geometry of three such models with variational autoencoder on two different datasets. The proposed study shows that the distances and interpolations in the latent space are significantly improved with Riemannian metric owing to the curvature of the space.
AB - Deep generative models like variational autoencoders approximate the intrinsic geometry of high dimensional data manifolds by learning a set of low-dimensional latent-space variables and an embedding function. The geometrical properties of these latent spaces has been studied under the lens of Riemannian geometry; via analysis of the non-linearity of the generator function. In new developments, deep generative models have been used for learning semantically meaningful 'disentangled' representations; that capture task relevant attributes while being invariant to other attributes. In this work, we explore the geometry of popular generative models for disentangled representation learning. We use several metrics to compare the properties of latent spaces of disentangled representation models in terms of class separability and curvature of the latent-space. The proposed study establishes that the class distinguishable features in the disentangled latent space exhibits higher curvature as opposed to a variational autoencoder. We evaluate and compare the geometry of three such models with variational autoencoder on two different datasets. The proposed study shows that the distances and interpolations in the latent space are significantly improved with Riemannian metric owing to the curvature of the space.
KW - Deep generative models
KW - Disentangled representations
KW - Riemannian Geometry
UR - http://www.scopus.com/inward/record.url?scp=85098111953&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85098111953&partnerID=8YFLogxK
U2 - 10.1145/3293353.3293422
DO - 10.1145/3293353.3293422
M3 - Conference contribution
AN - SCOPUS:85098111953
T3 - ACM International Conference Proceeding Series
BT - Proceedings - 11th Indian Conference on Computer Vision, Graphics and Image Processing, ICVGIP 2018
PB - Association for Computing Machinery
T2 - 11th Indian Conference on Computer Vision, Graphics and Image Processing, ICVGIP 2018
Y2 - 18 December 2018 through 22 December 2018
ER -