mlsanityData to 14 Oct 2026

FONDUE: an algorithm to find the optimal dimensionality of the latent representations of variational autoencoders

arXiv 2209.12806 ↗First posted 26 Sep 2022cs.LGcited by 5

Not on the map. Why?

Abstract (arXiv)

When training a variational autoencoder (VAE) on a given dataset, determining the optimal number of latent variables is mostly done by grid search: a costly process in terms of computational time and carbon footprint. In this paper, we explore the intrinsic dimension estimation (IDE) of the data and latent representations learned by VAEs. We show that the discrepancies between the IDE of the mean and sampled representations of a VAE after only a few steps of training reveal the presence of passive variables in the latent space, which, in well-behaved VAEs, indicates a superfluous number of dimensions. Using this property, we propose FONDUE: an algorithm which quickly finds the number of latent dimensions after which the mean and sampled representations start to diverge (i.e., when passive variables are introduced), providing a principled method for selecting the number of latent dimensions for VAEs and autoencoders.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 23 of its 36 references that are in the map. First posted on arXiv 26 Sep 2022.

Builds on

23 of 36 references.

References · 23

  1. 2013Auto-Encoding Variational Bayescited by 4,053
  2. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  3. 2015Variational Inference with Normalizing Flowscited by 2,170
  4. 2018Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representationscited by 941
  5. 2016Tutorial on Variational Autoencoderscited by 409
  6. 2021The Intrinsic Dimension of Images and Its Impact on Learningcited by 315
  7. 2018On the Optimization of Deep Networks: Implicit Acceleration by Overparameterizationcited by 284
  8. 2019Intrinsic dimension of data representations in deep neural networkscited by 282
  9. 2018Estimating the intrinsic dimension of datasets by a minimal neighborhood informationcited by 250
  10. 2019Diagnosing and Enhancing VAE Modelscited by 193
  11. 2017Latent Space Oddity: on the Curvature of Deep Generative Modelscited by 174
  12. 2019Don't Blame the ELBO! A Linear VAE Perspective on Posterior Collapsecited by 121
11 more
  1. 2020Topology of deep neural networkscited by 112
  2. 2018Variational Autoencoders Pursue PCA Directions (by Accident)cited by 85
  3. 2018Neural Persistence: A Complexity Measure for Deep Neural Networks Using Algebraic Topologycited by 73
  4. 2018Explorations in Homeomorphic Variational Auto-Encodingcited by 65
  5. 2019Universality and individuality in neural dynamics across large populations of recurrent networkscited by 61
  6. 2018Geometry Score: A Method For Comparing Generative Adversarial Networkscited by 57
  7. 2018On the Intrinsic Dimensionality of Image Representationscited by 57
  8. 2019The Usual Suspects? Reassessing Blame for VAE Posterior Collapsecited by 52
  9. 2019The Impact of Neural Network Overparameterization on Gradient Confusion and Stochastic Gradient Descentcited by 46
  10. 2020Evaluating the Disentanglement of Deep Generative Models through Manifold Topologycited by 20
  11. 2021Be More Active! Understanding the Differences between Mean and Sampled Representations of Variational Autoencoderscited by 7

Built on by

Cited by year

Citing papers per year: 2023: 2; 2024: 3.

5 papers.

  1. 2023Deep Backtracking Counterfactuals for Causally Compliant Explanationscited by 9
  2. 2024Doubly Robust Conditional Independence Testing with Generative Neural Networkscited by 7
  3. 2024Analyzing Deep Transformer Models for Time Series Forecasting via Manifold Learningcited by 5
  4. 2023How good are variational autoencoders at transfer learning?cited by 3
  5. 2024Intrinsic Dimensionality of Fermi-Pasta-Ulam-Tsingou High-Dimensional Trajectories Through Manifold Learning: A Linear Approachcited by 1