Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Gibbs Inference for Statistical Model Estimation from Incomplete Data
Vaidotas Simkus, Benjamin Rhodes, Michael Gutmann
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 |
|---|---|---|---|
| Fit factor analysis with VGI instead of empirical-sample imputation | Factor analysis model with incomplete data · Toy data and FA-Frey | KL divergence to ground-truth model | Improved |
| Use VGI instead of MICE for fitting the FA model | Factor analysis model · Toy data (6-dim) | KL divergence to ground-truth model | No clear effect |
| Use VGI instead of MICE for fitting the FA model | Factor analysis model · FA-Frey (560-dim) | KL divergence to ground-truth model | Improved |
| Compare stochastic VGI/MICE with exact EM as missingness grows | VGI and MICE · FA-Frey | KL divergence to ground-truth model | Hurt |
| Increase number of incomplete training points | VGI on toy FA model · Toy data, consistency | RMSE of parameters and KL divergence | Improved |
| Increase fraction of missing data | Variational conditionals · Toy data | KL divergence of variational conditionals to ground truth | Hurt |
| Use partially shared network for variational conditionals instead of independent networks (ablation) | Variational conditionals · Toy data and FA-Frey | median KL to fitted-model conditionals vs missingness | Improved |
| Use shared variational model instead of independent models (ablation) | VGI · Toy data | accuracy of fitted FA model | No clear effect |
| Use extended-Gibbs variational conditionals instead of standard conditionals (ablation) | VGI · Toy data (low dimension) | KL of conditionals and fitted model accuracy | Hurt |
| Use extended-Gibbs variational conditionals instead of standard conditionals (ablation) | VGI · Higher-dimensional data | VGI estimation accuracy | No clear effect |
| Use Adam instead of AMSGrad for variational parameters (ablation) | VGI · FA experiments | training stability | Hurt |
| Stack K imputations with 1/K weights instead of parameter averaging after Procrustes rotation | MICE baseline for FA · Toy FA | KL divergence to ground truth | No clear effect |
Closest papers by meaning
- On the properties of variational approximations of Gibbs posteriors (JMLR 2016 · similarity 0.85)
- Herded Gibbs Sampling (JMLR 2016 · similarity 0.83)
- Parameter Estimation in DAGs from Incomplete Data via Optimal Transport (ICML 2024 · similarity 0.82)
- Sparse Markov Models for High-dimensional Inference (JMLR 2023 · similarity 0.82)
- Variational Inference in high-dimensional linear regression (JMLR 2022 · similarity 0.82)
- Score Matching with Missing Data (ICML 2025 · 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.
