mlsanityData to 14 Oct 2026

Inference Suboptimality in Variational Autoencoders

arXiv 1801.03558 ↗First posted 10 Jan 2018cs.LG stat.MLcited by 132

Not on the map. Why?

Abstract (arXiv)

Amortized inference allows latent-variable models trained via variational learning to scale to large datasets. The quality of approximate inference is determined by two factors: a) the capacity of the variational distribution to match the true posterior and b) the ability of the recognition network to produce good variational parameters for each datapoint. We examine approximate inference in variational autoencoders in terms of these factors. We find that divergence from the true posterior is often due to imperfect recognition networks, rather than the limited complexity of the approximating distribution. We show that this is due partly to the generator learning to accommodate the choice of approximation. Furthermore, we show that the parameters used to increase the expressiveness of the approximation play a role in generalizing inference rather than simply improving the complexity of the approximation.

Key facts. Cited by 132 papers in the map (data through 14 Oct 2026). First posted on arXiv 10 Jan 2018.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 27; 2022: 26; 2023: 24; 2024: 20; 2025: 17; 2026: 18.

132 papers.

Sort:

By area:

  1. 2021Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Modelscited by 252
  2. 2021COIN: COmpression with Implicit Neural representationscited by 233
  3. 2021NeRF-VAE: A Geometry Aware 3D Scene Generative Modelcited by 112
  4. 2021Deep Adaptive Design: Amortizing Sequential Bayesian Experimental Designcited by 106
  5. 2023ATT3D: Amortized Text-to-3D Object Synthesiscited by 87
  6. 2022Tutorial on amortized optimizationcited by 83
  7. 2022An Introduction to Neural Data Compressioncited by 79
  8. 2021Hierarchical VAEs Know What They Don't Knowcited by 58
  9. 2024LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesiscited by 39
  10. 2024Neural Methods for Amortized Inferencecited by 39
  11. 2021Overfitting for Fun and Profit: Instance-Adaptive Data Compressioncited by 38
  12. 2023Amortized Variational Inference: When and Why?cited by 29
38 more
  1. 2021Data Augmentation in High Dimensional Low Sample Size Setting Using a Geometry-Based Variational Autoencodercited by 27
  2. 2022Hybrid Predictive Coding: Inferring, Fast and Slowcited by 26
  3. 2022Pythae: Unifying Generative Autoencoders in Python -- A Benchmarking Use Casecited by 26
  4. 2021Learning Opinion Summarizers by Selecting Informative Reviewscited by 25
  5. 2021How Tight Can PAC-Bayes be in the Small Data Regime?cited by 24
  6. 2022A Geometric Perspective on Variational Autoencoderscited by 24
  7. 2023Computationally-Efficient Neural Image Compression with Shallow Decoderscited by 23
  8. 2022Flexible Neural Image Compression via Code Editingcited by 20
  9. 2022Missing Data Imputation and Acquisition with Deep Hierarchical Models and Hamiltonian Monte Carlocited by 19
  10. 2021Instance-Adaptive Video Compression: Improving Neural Codecs by Training on the Test Setcited by 18
  11. 2022Amortized Variational Inference: A Systematic Reviewcited by 16
  12. 2021Variational Multi-Task Learning with Gumbel-Softmax Priorscited by 15
  13. 2021Mind the Gap when Conditioning Amortised Inference in Sequential Latent-Variable Modelscited by 14
  14. 2022Model-agnostic out-of-distribution detection using combined statistical testscited by 14
  15. 2023Learnable Topological Features for Phylogenetic Inference via Graph Neural Networkscited by 14
  16. 2021Motion Planning by Learning the Solution Manifold in Trajectory Optimizationcited by 13
  17. 2024An improved tabular data generator with VAE-GMM integrationcited by 13
  18. 2021VAE Approximation Error: ELBO and Exponential Familiescited by 12
  19. 2023A Deep Learning Method for Comparing Bayesian Hierarchical Modelscited by 12
  20. 2023Revisiting Structured Variational Autoencoderscited by 11
  21. 2024ASPIRE: Iterative Amortized Posterior Inference for Bayesian Inverse Problemscited by 11
  22. 2024Poisson Variational Autoencodercited by 11
  23. 2022Learning Neural Set Functions Under the Optimal Subset Oraclecited by 10
  24. 2022Alleviating Adversarial Attacks on Variational Autoencoders with MCMCcited by 10
  25. 2022Generalization Gap in Amortized Inferencecited by 10
  26. 2021Regularization-Agnostic Compressed Sensing MRI Reconstruction with Hypernetworkscited by 9
  27. 2022Bit Allocation using Optimizationcited by 9
  28. 2023Fully Bayesian Autoencoders with Latent Sparse Gaussian Processescited by 9
  29. 2022Toward Adaptive Semantic Communications: Efficient Data Transmission via Online Learned Nonlinear Transform Source-Channel Codingcited by 8
  30. 2025Bridge the Inference Gaps of Neural Processes via Expectation Maximizationcited by 8
  31. 2022Reducing The Amortization Gap of Entropy Bottleneck In End-to-End Image Compressioncited by 7
  32. 2022Variational Laplace Autoencoderscited by 7
  33. 2024Sequential Monte Carlo for Inclusive KL Minimization in Amortized Variational Inferencecited by 7
  34. 2021Reducing the Amortization Gap in Variational Autoencoders: A Bayesian Random Function Approachcited by 6
  35. 2021Generative Text Modeling through Short Run Inferencecited by 6
  36. 2022Improving VAE-based Representation Learningcited by 6
  37. 2023Diverse and Faithful Knowledge-Grounded Dialogue Generation via Sequential Posterior Inferencecited by 6
  38. 2024Domain Generalization with Small Datacited by 6

50 most cited of 132 All 132 (JSON)