Research map / Probabilistic and scientific modeling / Variational and generative sampling
Neural Posterior Estimation with Latent Basis Expansions
Declan McNamara, Yicun Duan, Jeffrey Regier
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 LBF-NPE instead of MDN | NPE variational family · 2D posterior: Bands | forward KL: 0.182 -> 0.0048 | Improved |
| use LBF-NPE instead of neural spline flow | NPE variational family · 2D posterior: Ring | forward KL: 0.017 -> 0.0054 | Improved |
| use LBF-NPE instead of RealNVP | NPE variational family · 2D posterior: Spiral | reverse KL: 0.666 -> 0.188 | Improved |
| use LBF-NPE instead of NSF | NPE variational family · 2D posterior: Spiral | negative log-likelihood: 0.838 -> 0.727 | Hurt |
| use LBF-NPE instead of MDN | NPE variational family · DC2 redshift estimation | held-out NLL: -50,648 -> -57,220 | Improved |
| use LBF-NPE instead of NSF | NPE variational family · DC2 redshift estimation | held-out NLL: -55,389 -> -57,220 | Improved |
| use fixed B-spline basis (convex objective) instead of MDN | NPE · Sinusoidal toy, 20 seeds | convergence to optimum | Improved |
| increase number of MDN components beyond five (ablation) | MDN · DC2 redshift estimation | NLL | No clear effect |
| increase number of coupling layers in normalizing flow beyond 16 (ablation) | NSF baseline · 2D posterior: Spiral | fit quality | No clear effect |
| use K = 2 basis functions instead of 20 (ablation) | LBF-NPE · 2D posterior: Bands | posterior fit | No clear effect |
| add stereographic projection reparameterization (ablation) | LBF-NPE adaptive basis · 2D posteriors | identifiability and training | Improved |
Closest papers by meaning
- Cross-Fitting for Neural Posterior Estimation (NeurIPS 2026 · similarity 0.89)
- Efficient Low Rank Gaussian Variational Inference for Neural Networks (NeurIPS 2020 · similarity 0.89)
- Globally Convergent Variational Inference (NeurIPS 2024 · similarity 0.89)
- The k-tied Normal Distribution: A Compact Parameterization of Gaussian Mean Field Posteriors in Bayesian Neural Networks (ICML 2020 · similarity 0.89)
- Bayesian Basis Function Approximation for Scalable Gaussian Process Priors in Deep Generative Models (ICML 2025 · similarity 0.88)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.88)
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.
