Research map / Probabilistic and scientific modeling / Variational and generative sampling
SUMO: Unbiased Estimation of Log Marginal Probability for Latent Variable 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 IWAE training with SUMO at expected cost k=5 | latent variable model · MNIST | test negative log-likelihood: 85.28$\pm$0.01 -> 85.09$\pm$0.01 | Helped |
| Replace IWAE training with SUMO at expected cost k=15 | latent variable model · MNIST | test negative log-likelihood: 84.89$\pm$0.03 -> 84.71$\pm$0.02 | Helped |
| Replace IWAE training with SUMO at expected cost k=50 | latent variable model · MNIST | test negative log-likelihood: 84.50$\pm$0.02 -> 84.40$\pm$0.03 | Helped |
| Replace IWAE training with SUMO at expected cost k=5 | latent variable model · OMNIGLOT | test negative log-likelihood: 104.96$\pm$0.04 -> 104.85$\pm$0.04 | Helped |
| Replace IWAE training with SUMO at expected cost k=15 | latent variable model · OMNIGLOT | test negative log-likelihood: 104.53$\pm$0.05 -> 104.29$\pm$0.12 | Helped |
| Replace IWAE training with SUMO at expected cost k=50 | latent variable model · OMNIGLOT | test negative log-likelihood: 103.99$\pm$0.12 -> 103.79$\pm$0.14 | Helped |
| Use unbiased SUMO instead of lower-bound IWAE for reverse-KL training | latent variable model · Funnel target reverse KL | reverse KL optimization stability | Helped |
| Use SUMO rather than IWAE with entropy regularization | latent variable policy · Quadratic pseudo-Boolean optimization | reward optimization performance | Helped |
| Use SUMO instead of an autoregressive policy | latent variable policy · Quadratic pseudo-Boolean optimization, d=500 | training speed: 19.2× | Helped |
| Apply gradient clipping during SUMO density-model training (ablation) | SUMO · MNIST density modeling | training stability | Helped |
Closest papers by meaning
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.87)
- Efficient Marginalization of Discrete and Structured Latent Variables via Sparsity (NeurIPS 2020 · similarity 0.87)
- Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimator (ICLR 2021 · similarity 0.87)
- Variational Inference for Monte Carlo Objectives (ICML 2016 · similarity 0.87)
- Variational Bayes on Monte Carlo Steroids (NeurIPS 2016 · similarity 0.87)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 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.