Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
read the original abstract
Disentangled representations, where the higher level data generative factors are reflected in disjoint latent dimensions, offer several benefits such as ease of deriving invariant representations, transferability to other tasks, interpretability, etc. We consider the problem of unsupervised learning of disentangled representations from large pool of unlabeled observations, and propose a variational inference based approach to infer disentangled latent factors. We introduce a regularizer on the expectation of the approximate posterior over observed data that encourages the disentanglement. We also propose a new disentanglement metric which is better aligned with the qualitative disentanglement observed in the decoder's output. We empirically observe significant improvement over existing methods in terms of both disentanglement and data likelihood (reconstruction quality).
This paper has not been read by Pith yet.
Forward citations
Cited by 3 Pith papers
-
Unsupervised Disentanglement Without Compromises : How Functional Orthogonality Enforces Identifiability
Enforcing local orthogonality on the Jacobian of the generative mapping yields identifiability for general nonlinear models when the latent domain has full combinatorial support.
-
Monotonic Kolmogorov-Arnold Networks: A Theoretical and Empirical Study of Monotonicity as an Inductive Bias
MKAN adds hard unconstrained monotonicity to KANs via reparameterization and proves a size bound of at most 2N* for monotone equivalents of ball-partition feature extractors.
-
MB-Loc: Multi-planar Bird's-eye-view Localization in outdoor LiDAR scenes
MB-Loc projects LiDAR point clouds into multi-planar BEV images, applies 2D CNNs with a KL-regularized latent bottleneck and 3D augmentations, and reports real-time state-of-the-art localization accuracy on the NCLT dataset.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.