We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Indeterminacy and Strong Identifiability in Generative Models

Abstract: Most modern probabilistic generative models, such as the variational autoencoder (VAE), have certain indeterminacies that are unresolvable even with an infinite amount of data. Different tasks tolerate different indeterminacies, however recent applications have indicated the need for strongly identifiable models, in which an observation corresponds to a unique latent code. Progress has been made towards reducing model indeterminacies while maintaining flexibility, and recent work excludes many--but not all--indeterminacies. In this work, we motivate model-identifiability in terms of task-identifiability, then construct a theoretical framework for analyzing the indeterminacies of latent variable models, which enables their precise characterization in terms of the generator function and prior distribution spaces. We reveal that strong identifiability is possible even with highly flexible nonlinear generators, and give two such examples. One is a straightforward modification of iVAE (arXiv:1907.04809 [stat.ML]); the other uses triangular monotonic maps, leading to novel connections between optimal transport and identifiability.
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG)
Cite as: arXiv:2206.00801 [stat.ML]
  (or arXiv:2206.00801v3 [stat.ML] for this version)

Submission history

From: Quanhan Xi [view email]
[v1] Thu, 2 Jun 2022 00:01:27 GMT (81kb)
[v2] Mon, 20 Jun 2022 06:54:21 GMT (84kb)
[v3] Fri, 23 Dec 2022 21:23:28 GMT (108kb,D)

Link back to: arXiv, form interface, contact.