Research map / Probabilistic and scientific modeling / Variational and generative sampling
Model evidence from nonequilibrium simulations
Michael Habeck
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 BAR instead of forward AIS | BAR · Ising 32x32 | log Z estimate (true 1339.27): 1333.66 -> 1338.05 | Improved |
| use histogram estimator instead of forward AIS | Histogram estimator · Ising 32x32 | log Z estimate: 1333.66 -> 1338.28 | Improved |
| use reverse AIS instead of BAR | Reverse AIS · Ising 32x32 | log Z estimate: 1342.05 | Hurt |
| raise equilibration steps N (ablation) | BAR and histogram estimator · Ising 32x32 | evidence accuracy | Improved |
| combine forward and reverse simulations (combined cumulant) rather than forward only (ablation) | Cumulant estimator · Gaussian toy model | bias | Improved |
| reverse AIS instead of forward AIS (ablation) | AIS · Gaussian toy model | accuracy | Improved |
| increase annealing steps K at fixed budget K x N = 10^9 (ablation) | BAR and histogram · Potts model | log Z accuracy | No clear effect |
| more equilibration steps N (ablation) | Histogram estimator Gibbs version · Potts model | posterior of log Z | Improved |
| anneal marginal model p(h) instead of full model p(v,h) (ablation) | Annealing · RBM MNIST | log Z estimate: 451.43 -> 451.48 | No clear effect |
| anneal marginal model instead of full model (ablation) | Annealing · RBM MNIST | speed | Improved |
| increase Gibbs steps 10 -> 100 in thermal annealing (ablation) | Thermal annealing · Gaussian mixture | log evidence: -259.72 ± 0.60 -> -259.47 ± 0.36 | Improved |
| increase Gibbs steps 10 -> 100 in data annealing (ablation) | Data annealing · Gaussian mixture | log evidence: –257.52 ± 0.97 -> -258.32 ± 1.21 | Improved |
Closest papers by meaning
- Measuring the reliability of MCMC inference with bidirectional Monte Carlo (NeurIPS 2016 · similarity 0.89)
- Likelihood-free MCMC with Amortized Approximate Ratio Estimators (ICML 2020 · similarity 0.87)
- Surrogate Likelihoods for Variational Annealed Importance Sampling (ICML 2022 · similarity 0.87)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.85)
- On Estimating the Gradient of the Expected Information Gain in Bayesian Experimental Design (AAAI 2024 · similarity 0.85)
- Learning Likelihood-Free Reference Priors (ICML 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.
