Research map / Probabilistic and scientific modeling / Variational and generative sampling
MCMC Variational Inference via Uncorrected Hamiltonian Annealing
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 UHA instead of IW | VAE with one-hidden-layer architecture · MNIST test, K=64 | ELBO: -89.0 -> -87.6 | Helped |
| Use UHA instead of IW | VAE with one-hidden-layer architecture · Letters test, K=64 | ELBO: -132.7 -> -130.9 | Helped |
| Use UHA instead of IW | VAE with one-hidden-layer architecture · KMNIST test, K=64 | ELBO: -177.0 -> -171.6 | Helped |
| Increase UHA bridging density count from K=8 to K=64 (ablation) | VAE with one-hidden-layer architecture · MNIST test | ELBO: -89.8 -> -87.6 | Helped |
| Increase UHA bridging density count from K=8 to K=64 (ablation) | VAE with one-hidden-layer architecture · Letters test | ELBO: -133.5 -> -130.9 | Helped |
| Increase UHA bridging density count from K=8 to K=64 (ablation) | VAE with one-hidden-layer architecture · KMNIST test | ELBO: -176.6 -> -171.6 | Helped |
| Use UHA instead of IW | Student-t target inference · Dimension 20, K=128 | ELBO: -0.14 vs -0.14 | No clear effect |
| Use UHA instead of IW | Student-t target inference · Dimension 500; UHA K=16 versus IW K=1024 | ELBO: -12.0 -> -9.0 | Helped |
| Tune all reported UHA parameters rather than only epsilon and eta (ablation) | UHA Bayesian inference · Logistic regression models, K=64 compared with HAIS K=512 | ELBO | Helped |
| Tune bridging coefficients beta and initial distribution q(z) (ablation) | UHA Bayesian inference · Four inference models, K=64 | ELBO | Helped |
| Use UHA instead of IW | VAE with two-hidden-layer architecture · Test log-likelihood for smaller K | log-likelihood | Mixed |
| Increase UHA bridging density count (ablation) | VAE training · Some datasets at larger K | training stability | Hurt |
Closest papers by meaning
- Surrogate Likelihoods for Variational Annealed Importance Sampling (ICML 2022 · similarity 0.92)
- Differentiable Annealed Importance Sampling Minimizes The Jensen-Shannon Divergence Between Initial and Target Distribution (ICML 2024 · similarity 0.87)
- Provable Benefit of Annealed Langevin Monte Carlo for Non-log-concave Sampling (ICLR 2025 · similarity 0.87)
- Revisiting Unbiased Implicit Variational Inference (ICML 2025 · similarity 0.87)
- Reparameterized Importance Sampling for Robust Variational Bayesian Neural Networks (ICML 2024 · similarity 0.87)
- Annealing in variational inference mitigates mode collapse: a theoretical study on Gaussian mixtures (NeurIPS 2026 · 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.