Research map / Probabilistic and scientific modeling / Variational and generative sampling
MIWAE: Deep Generative Modelling and Imputation of Incomplete Data Sets
Pierre-Alexandre Mattei, Jes Frellsen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MIWAE multiple imputation instead of complete training data | Convolutional classifier · MNIST static binarisation, 50% pixels missing | Test accuracy: 0.9866 (0.0007) -> 0.9868 (0.0008) | Improved |
| Use MIWAE multiple imputation instead of complete training data | Convolutional classifier · MNIST static binarisation, 50% pixels missing | Test cross-entropy: 0.0464 (0.0026) -> 0.0509 (0.0044) | Hurt |
| Use MIWAE single imputation instead of missForest imputation | Convolutional classifier · MNIST static binarisation, 50% pixels missing | Test accuracy: 0.9805 (0.0018) -> 0.9847 (0.0009) | Improved |
| Use MIWAE single imputation instead of missForest imputation | Convolutional classifier · MNIST static binarisation, 50% pixels missing | Test cross-entropy: 0.0645 (0.0066) -> 0.0510 (0.0035) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · Banknote | Mean-squared error: 0.593 (0.059) -> 0.446 (0.038) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · Breast | Mean-squared error: 0.318 (0.018) -> 0.280 (0.021) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · Concrete | Mean-squared error: 0.587(0.026) -> 0.501 (0.040) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · Red | Mean-squared error: 0.686 (0.120) -> 0.643 (0.026) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · White | Mean-squared error: 0.782 (0.018) -> 0.735 (0.033) | Improved |
| Use MIWAE with K=20 instead of MVAE with K=1 (ablation) | MIWAE single imputation · Yeast | Mean-squared error: 0.997 (0.064) -> 0.964(0.057) | Improved |
| Use zero imputation instead of oracle imputation during training (ablation) | MIWAE with K=50 · MNIST static binarisation, 50% pixels missing | Test log-likelihood | No clear effect |
| Use zero imputation instead of oracle imputation during training (ablation) | MVAE with K=1 · MNIST static binarisation, 50% pixels missing | Test log-likelihood | Hurt |
Later papers that compared with this method
4 later accepted papers report MIWAE as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- AdaVAE: Bayesian Structural Adaptation for Variational Autoencoders (NeurIPS 2023 · better on 4 of 5 shared results)
- Missing Data Imputation and Acquisition with Deep Hierarchical Models and Hamiltonian Monte Carlo (NeurIPS 2022 · better on 6 of 6 shared results)
- Projected Latent Markov Chain Monte Carlo: Conditional Sampling of Normalizing Flows (ICLR 2021 · better on 0 of 6 shared results)
- Tomographic Auto-Encoder: Unsupervised Bayesian Recovery of Corrupted Data (ICLR 2021 · better on 6 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DiffPuter: Empowering Diffusion Models for Missing Data Imputation (ICLR 2025 · similarity 0.88)
- Distribution Shift in Missing Data Imputation: A Risk-Based Perspective and Importance-Weighted Correction under MAR (NeurIPS 2026 · similarity 0.86)
- Missing Data Imputation by Reducing Mutual Information with Rectified Flows (NeurIPS 2025 · similarity 0.86)
- Learning What Not to Impute: An Uncertainty-Aware Diffusion Framework for Meaningful Missingness (NeurIPS 2026 · similarity 0.85)
- To Predict or Not to Predict? Proportionally Masked Autoencoders for Tabular Data Imputation (AAAI 2025 · similarity 0.85)
- Rethinking the Diffusion Models for Missing Data Imputation: A Gradient Flow Perspective (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
