Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Tractable Probabilistic Models from Inconsistent Local Estimates
Shasha Jin, Vasundhara Komaragiri, Tahrima Rahman, Vibhav Giridhar Gogate
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 |
|---|---|---|---|
| Refine cutset network Q with inconsistent local pairwise estimates (Algorithm 1) | Chow-Liu tree · 20 benchmark datasets avg, 0% evidence | negative cross-entropy: -79.81 -> -64.03 | Improved |
| Refine Q with local estimates | cutset network (CN) · 20 datasets avg, 0% evidence | negative cross-entropy: -75.14 -> -59.01 | Improved |
| Refine Q with local estimates | mixture of cutset networks · 20 datasets avg, 0% evidence | negative cross-entropy: -69.67 -> -55.43 | Improved |
| Refine Q with local estimates | CLT · 20 datasets avg, 50% evidence | negative cross-entropy: -61.67 -> -50.28 | Improved |
| Refine Q with local estimates | MCN · 20 datasets avg, 80% evidence | negative cross-entropy: -39.19 -> -30.19 | Improved |
| Refine Q with local estimates | MCN · movie, 0% evidence | negative cross-entropy: -64.99 -> -34.14 | Improved |
| Refine Q with local estimates | CN · nltcs-like small dataset msnbc, 0% evidence | negative cross-entropy: -6.99 -> -6.36 | Improved |
| Refine Q with local estimates | MCN · msnbc, 50% evidence | negative cross-entropy: -5.03 -> -2.14 | Improved |
| Increase perturb rate h of initial network Q (ablation) | Q vs R · movie (Figure 2) | negative cross-entropy | Mixed |
| Increase noise std sigma on local statistics (ablation) | R · pumsb* (Figure 3) | negative cross-entropy | Hurt |
| Use sigma above 0.2 for local-statistic noise (ablation) | R vs Q · pumsb* | negative cross-entropy | Mixed |
| Increase number of evidence variables from 50% to 80% (ablation) | Algorithm 1 · 20 datasets avg | relative improvement of R over Q | No clear effect |
Closest papers by meaning
- Look Ma, No Latent Variables: Accurate Cutset Networks via Compilation (ICML 2019 · similarity 0.89)
- Continuous Mixtures of Tractable Probabilistic Models (AAAI 2023 · similarity 0.87)
- The Sum-Product Theorem: A Foundation for Learning Tractable Models (ICML 2016 · similarity 0.87)
- On the Hardness of Approximating Distributions with Tractable Probabilistic Models (NeurIPS 2025 · similarity 0.87)
- Learning Ensembles of Cutset Networks (AAAI 2016 · similarity 0.87)
- Learning Local Neighborhoods of Non-Gaussian Graphical Models (AAAI 2025 · similarity 0.85)
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.
