Research map / Architectures and efficiency / RNNs and normalization
Processing of missing data by neural networks
Marek Śmieja, Łukasz Struski, Jacek Tabor, Bartosz Zieliński, Przemysław Spurek
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 |
|---|---|---|---|
| Represent missing inputs with a learned density and use expected first-layer neuron activations instead of single-value imputation | Autoencoder on MNIST incomplete images · MNIST incomplete images | total reconstruction MSE: 0.01189 -> 0.01056 | Improved |
| Represent missing inputs with expected activations rather than k-nn imputation | Autoencoder on MNIST incomplete images · MNIST incomplete images | MSE inside mask: 0.00722 -> 0.00810 | Hurt |
| Represent missing inputs with expected activations rather than k-nn imputation | Autoencoder on MNIST incomplete images · MNIST incomplete images | MSE outside mask: 0.00468 -> 0.00246 | Improved |
| Use expected activations instead of MICE imputation | MLP on ESR · ESR, 25% missing | accuracy: 0.823 -> 0.815 | Hurt |
| Use expected activations instead of MICE imputation | MLP on ESR · ESR, 50% missing | accuracy: 0.816 -> 0.817 | Improved |
| Use expected activations instead of MICE imputation | MLP on ESR · ESR, 75% missing | accuracy: 0.786 -> 0.787 | No clear effect |
| Use expected activations instead of MICE imputation | MLP on ESR · ESR, 90% missing | accuracy: 0.670 -> 0.760 | Improved |
| Use expected activations rather than the strongest reported baseline | RBFN on UCI datasets with internal missing attributes · hepatitis | accuracy: 0.825 -> 0.846 | Improved |
| Use expected activations rather than the strongest reported baseline | RBFN on UCI datasets with internal missing attributes · kidney disease | accuracy: 0.995 -> 0.993 | Hurt |
| Use expected activations rather than the strongest reported baseline | RBFN on UCI datasets with internal missing attributes · horse | accuracy: 0.826 -> 0.864 | Improved |
| Use expected activations rather than the strongest reported baseline | RBFN on UCI datasets with internal missing attributes · pima | accuracy: 0.769 -> 0.747 | Hurt |
| Use expected activations rather than the strongest reported baseline | RBFN on UCI datasets with internal missing attributes · winconsin | accuracy: 0.970 -> 0.970 | No clear effect |
Closest papers by meaning
- NeuMiss networks: differentiable programming for supervised learning with missing values. (NeurIPS 2020 · similarity 0.86)
- How to deal with missing data in supervised deep learning? (ICLR 2022 · similarity 0.86)
- Missingness-Pattern-Adaptive Learning With Incomplete Data (TPAMI 2023 · similarity 0.86)
- Missing Data Imputation by Reducing Mutual Information with Rectified Flows (NeurIPS 2025 · similarity 0.85)
- Learning What Not to Impute: An Uncertainty-Aware Diffusion Framework for Meaningful Missingness (NeurIPS 2026 · similarity 0.82)
- Distribution Shift in Missing Data Imputation: A Risk-Based Perspective and Importance-Weighted Correction under MAR (NeurIPS 2026 · similarity 0.82)
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.
