Research map / Probabilistic and scientific modeling / Variational and generative sampling
Scaling Up Probabilistic Circuits by Latent Variable Distillation
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 |
|---|---|---|---|
| add latent variable distillation (LVD) to HCLT-based patch PC | large PCs on ImageNet32 · ImageNet32 | test bpd: 4.82 -> 4.39 | Helped |
| add latent variable distillation (LVD) to HCLT-based patch PC | large PCs on ImageNet64 · ImageNet64 | test bpd: 4.67 -> 4.12 | Helped |
| add latent variable distillation (LVD) to HCLT-based patch PC | large PCs on CIFAR · CIFAR | test bpd: 4.61 -> 4.38 | Helped |
| add latent variable distillation to RAT-SPN structure (ablation) | RAT-SPN on ImageNet32 · ImageNet32 | test bpd: 6.90 -> 4.99 | Helped |
| use VQ-VAE codes instead of MAE for LV assignments (ablation) | LVD image PC · ImageNet32 | test bpd: 4.44 vs 4.39 (MAE) | Hurt |
| change MAE patch size from 4 to 2 (ablation) | LVD image PC · ImageNet32 | test bpd: 4.39 -> 4.57 | Hurt |
| change MAE patch size from 4 to 8 (ablation) | LVD image PC · ImageNet32 | test bpd: 4.39 -> 4.44 | Hurt |
| add latent distribution training step before full finetuning | LVD training pipeline · ImageNet32 | convergence speed | Helped |
| add LVD (no finetuning effect quantified) on large PCs; LVD bpd decreases with model size whereas EM plateaus | PC parameter scaling · TPM image benchmarks | test bpd vs parameter count | Helped |
Closest papers by meaning
- The Expressivity Boundary of Probabilistic Circuits: A Comparison with Large Language Models (NeurIPS 2026 · similarity 0.90)
- Scaling Tractable Probabilistic Circuits: A Systems Perspective (ICML 2024 · similarity 0.90)
- Scaling Continuous Latent Variable Models as Probabilistic Integral Circuits (NeurIPS 2024 · similarity 0.89)
- Scaling Probabilistic Circuits via Monarch Matrices (ICML 2025 · similarity 0.89)
- On the Hardness of Approximating Distributions with Tractable Probabilistic Models (NeurIPS 2025 · similarity 0.87)
- Sum of Squares Circuits (AAAI 2025 · similarity 0.87)
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.