Research map / Probabilistic and scientific modeling / Variational and generative sampling
Amortized Population Gibbs Samplers with Neural Sufficient Statistics
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 |
|---|---|---|---|
| Use APG with learned conditional block proposals and K=20 sweeps instead of HMC-RWS with LF=10 | GMM inference · GMM test corpus | log pθ(x, z): -4.673 -> -4.469 | Helped |
| Use APG with learned conditional block proposals and K=20 sweeps instead of HMC-RWS with LF=10 | DMM inference · DMM test corpus | log pθ(x, z): -2.184 -> -2.035 | Helped |
| Use APG with learned conditional block proposals and K=20 sweeps instead of HMC-RWS with LF=10 | Bouncing MNIST inference · BC-MNIST test corpus | log pθ(x, z): -1.303 -> -0.620 | Helped |
| Use APG with K=20 instead of HMC-RWS with LF=10 | Bouncing MNIST reconstruction · D=5, T=20; D=5, T=100; D=3, T=20; D=3, T=100 | mean-squared error: 261.2 -> 103.8; 267.9 -> 102.3; 124.1 -> 58.9; 127.4 -> 55.7 | Helped |
| Increase APG sweeps from K=10 to K=20 (ablation) | APG sampler · GMM, DMM, and BC-MNIST test corpora | log pθ(x, z): -4.556 -> -4.469; -2.040 -> -2.035; -0.652 -> -0.620 | Helped |
| Increase APG sweeps from K=10 to K=20 (ablation) | APG sampler · Bouncing MNIST reconstruction, four D/T settings | mean-squared error: 117.7 -> 103.8; 115.4 -> 102.3; 60.4 -> 58.9; 56.0 -> 55.7 | Helped |
| Use more sweeps with fewer particles under a fixed computation budget (ablation) | APG sampler · GMM | effective sample size (ESS) | Helped |
| Decompose the joint variables into smaller blocks rather than update them jointly (ablation) | APG sampler · GMM | effective sample size (ESS) | Helped |
| Learn conditional proposals instead of proposing from the prior (ablation) | Population Gibbs sampler · GMM | log pθ(x, z) | Helped |
Closest papers by meaning
- Robust Amortized Simulation-Based Inference via Learned Error Models (NeurIPS 2026 · similarity 0.88)
- Improved off-policy training of diffusion samplers (NeurIPS 2024 · similarity 0.87)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.86)
- Adaptive teachers for amortized samplers (ICLR 2025 · similarity 0.86)
- All-in-one simulation-based inference (ICML 2024 · similarity 0.86)
- Leveraging Self-Consistency for Data-Efficient Amortized Bayesian Inference (ICML 2024 · 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.