Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Deep Latent Gaussian Models with Markov Chain Monte Carlo
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 |
|---|---|---|---|
| increase number of HMC steps from 0 (standard VAE) to 20 during training (ablation) | HMC-DLGM · binarized MNIST | held-out log-likelihood: −90.14 -> −82.53 | Helped |
| use 2 HMC steps instead of 0 (ablation) | HMC-DLGM · binarized MNIST | held-out log-likelihood: −90.14 -> −85.15 | Helped |
| increase HMC steps from 2 to 20 (ablation) | HMC-DLGM · binarized MNIST | held-out log-likelihood: −85.15 -> −82.53 | Helped |
| HMC-DLGM-20 versus 1-layer IWAE | HMC-DLGM-20 · binarized MNIST | held-out log-likelihood: −84.78 -> −82.53 | Helped |
| HMC-DLGM-20 (1 layer) versus deeper multi-layer models | HMC-DLGM-20 · binarized MNIST | held-out log-likelihood: −82.53 -> −81.74 | Hurt |
| use many HMC steps instead of one HMC update | HMC-DLGM-2 vs HVI · binarized MNIST | held-out log-likelihood: −85.51 -> −85.15 | No clear effect |
| train with HMC instead of mean-field VAE | DLGM · MNIST samples | sample sharpness | Helped |
| train with HMC instead of mean-field VAE | DLGM · Jacobian singular value spectrum | number of used latent dimensions: 10 -> more gradual falloff | Helped |
| increase HMC steps to 20 (ablation) | HMC-DLGM · training cost | training time: 21 times longer | Hurt |
Closest papers by meaning
- Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors (ICML 2016 · similarity 0.89)
- Structured Stochastic Gradient MCMC (ICML 2022 · similarity 0.89)
- On the Anatomy of MCMC-Based Maximum Likelihood Learning of Energy-Based Models (AAAI 2020 · similarity 0.89)
- Beyond the Mean-Field: Structured Deep Gaussian Processes Improve the Predictive Uncertainties (NeurIPS 2020 · similarity 0.89)
- Learning Deep Latent Variable Models by Short-Run MCMC Inference With Optimal Transport Correction (CVPR 2021 · similarity 0.88)
- Online Variational Sequential Monte Carlo (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.