Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hierarchical Variational Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace mean-field variational family with hierarchical variational model (normalizing flow prior on mean-field parameters) | Deep exponential families (Poisson and Bernoulli) · New York Times and Science held-out perplexity | held-out perplexity (lower is better) | Mixed |
| Use HVM with multi-layer Poisson DEF instead of mean-field | Poisson DEF 100-30 and 100-30-15 · New York Times | held-out perplexity: 3896 -> 3396 and 3962 -> 3346 | Helped |
| Use HVM with normalizing flow prior instead of mean-field | Bernoulli DEF 100-30 on New York Times · New York Times | held-out perplexity: 3339 -> 3394 (mean-field better) | Hurt |
| Use HVM instead of mean-field on Science corpus across all six models | Deep exponential families on Science · Science held-out perplexity | held-out perplexity: mean improvement of about 180 points across all models reported in text | Helped |
| Place a mixture-of-Gaussians prior on Poisson rate parameters instead of mean-field | Correlated discrete Poisson latent variable model · Simulated two-dimensional discrete posterior | qualitative fidelity to true posterior | Helped |
Closest papers by meaning
- Importance Weighted Hierarchical Variational Inference (NeurIPS 2019 · similarity 0.91)
- Provably Scalable Black-Box Variational Inference with Structured Variational Families (ICML 2024 · similarity 0.90)
- Denoising Diffusion Variational Inference: Diffusion Models as Expressive Variational Posteriors (AAAI 2025 · similarity 0.88)
- Nearly Dimension-Independent Convergence of Mean-Field Black-Box Variational Inference (NeurIPS 2025 · similarity 0.88)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · similarity 0.88)
- Batch and match: black-box variational inference with a score-based divergence (ICML 2024 · 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.