Research map / Probabilistic and scientific modeling / Variational and generative sampling
Online but Accurate Inference for Latent Variable Models with Local Gibbs Sampling
Christophe Dupuy, Francis Bach
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 Gibbs local sampling in online EM (G-OEM) instead of variational online LDA | G-OEM · LDA on six datasets | held-out log-perplexity | Improved |
| use frequentist online EM instead of variational Bayesian (OLDA, SVB) | G-OEM, V-OEM++, SPLDA · LDA on all datasets | held-out log-perplexity | Improved |
| use Gibbs sampling instead of variational inference for local updates | G-OEM · LDA, all experiments | held-out log-perplexity | Improved |
| add inference boosting to online EM (ablation) | V-OEM++ · LDA, Synthetic and IMDB | log-perplexity | Improved |
| add inference boosting to online EM (ablation) | G-OEM++ · LDA | log-perplexity | No clear effect |
| use step-size 1/sqrt(t) instead of 1/t (ablation) | V-OEM++ vs SPLDA · LDA | log-perplexity | Improved |
| average iterates for kappa below 0.5 (ablation) | G-OEM · Synthetic and IMDB | perplexity | Improved |
| average iterates for kappa above 0.5 (ablation) | G-OEM · Synthetic and IMDB | perplexity | Hurt |
| increase internal updates from P=20 (1 pass) to P=200 (4 passes) (ablation) | OLDA · IMDB | log-perplexity: 681.9 -> 682.6 | No clear effect |
| increase internal updates from P=20 (1 pass) to P=200 (4 passes) (ablation) | SVB · IMDB | log-perplexity: 684.5 -> 683.8 | No clear effect |
| use fixed point update for alpha instead of gamma prior (ablation) | G-OEM · Synthetic, K=10 | perplexity | Improved |
| use G-OEM instead of VarGibbs | G-OEM · LDA and HDP | perplexity | Improved |
Closest papers by meaning
- Deep Attentive Variational Inference (ICLR 2022 · similarity 0.88)
- Predictive variational inference: Learn the predictively optimal posterior distribution (ICML 2026 · similarity 0.88)
- Latent Thought Models with Variational Bayes Inference-Time Computation (ICML 2025 · similarity 0.87)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.87)
- Large Language Bayes (NeurIPS 2025 · similarity 0.87)
- Denoising Diffusion Variational Inference: Diffusion Models as Expressive Variational Posteriors (AAAI 2025 · similarity 0.86)
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.
