Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Deep Latent Variable Models by Short-Run MCMC Inference With Optimal Transport Correction
Check 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 OT correction to short-run MCMC inference | deep latent variable model · MNIST, SVHN, CelebA | MSE: 0.015 -> 0.0008; 0.008 -> 0.002; 0.013 -> 0.010 | Helped |
| add OT correction to short-run MCMC inference | deep latent variable model · MNIST, SVHN, CelebA | FID: 19.21 -> 14.28; 29.44 -> 19.48; 37.87 -> 29.75 | Helped |
| add OT correction to short-run MCMC inference | deep latent variable model · MNIST anomaly detection, held-out digits 1, 4, 5, 7, 9 | AUPRC: 0.336± 0.008 -> 0.353±0.021; 0.630± 0.017 -> 0.770± 0.024; 0.619± 0.013 -> 0.726±0.030; 0.463± 0.009 -> 0.550±0.013; 0.4 | Helped |
| increase latent dimension from 20 to 40 to 64 (ablation) | proposed model on SVHN · SVHN | MSE: 0.011 -> 0.008 -> 0.002 | Helped |
| increase latent dimension from 20 to 40 to 64 (ablation) | proposed model on SVHN · SVHN | FID: 36.32 -> 24.73 -> 19.48 | Helped |
| apply OT correction (ablation) | proposed model with Langevin step size s=3e-2 · SVHN | FID: 22.70 -> 21.89 | Helped |
| apply OT correction (ablation) | proposed model with Langevin step size s=3e-2 · SVHN | MSE: 0.011 -> 0.013 | Hurt |
| use an excessively large Langevin step size s=6e-2 (ablation) | proposed model · SVHN | FID: 109.97 before -> 87.77 after | Hurt |
| increase Langevin steps from 30 to 50 to 100 (ablation) | proposed model · SVHN | MSE: 0.014 -> 0.011 -> 0.007 | Helped |
| increase Langevin steps from 30 to 50 to 100 (ablation) | proposed model · SVHN | FID: 22.32 -> 18.57 -> 15.43 | Helped |
| increase learning-step gradient-ascent iterations from 1 to 2 to 3 (ablation) | proposed model · SVHN | MSE: 0.013 -> 0.010 -> 0.008 | Helped |
| increase learning-step gradient-ascent iterations from 1 to 2 to 3 (ablation) | proposed model · SVHN | FID: 21.89 -> 17.32 -> 14.28 | Helped |
Closest papers by meaning
- Microcanonical Langevin Ensembles: Advancing the Sampling of Bayesian Neural Networks (ICLR 2025 · similarity 0.88)
- Variational Inference for SDEs Driven by Fractional Noise (ICLR 2024 · similarity 0.87)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.86)
- Online Variational Sequential Monte Carlo (ICML 2024 · similarity 0.86)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · similarity 0.86)
- SING: SDE Inference via Natural Gradients (NeurIPS 2025 · 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.