Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Deep Gaussian Markov Random Fields
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 a 5-layer DGMRF with 5×5 seq-filters rather than SPDE | DGMRF · Satellite temperatures test set | MAE: 1.10 -> 0.93 | Helped |
| Use a 5-layer DGMRF with 5×5 seq-filters rather than SPDE | DGMRF · Satellite temperatures test set | RMSE: 1.53 -> 1.25 | Helped |
| Use a 5-layer DGMRF with 5×5 seq-filters rather than SPDE | DGMRF · Satellite temperatures test set | CRPS: 0.83 -> 0.74 | Helped |
| Use a 5-layer DGMRF with 5×5 seq-filters rather than the best listed baseline | DGMRF · Satellite temperatures test set | INT: 7.44 -> 8.14 | Hurt |
| Use a 5-layer DGMRF with 5×5 seq-filters rather than NNGP | DGMRF · Satellite temperatures test set | CVG: 0.95 -> 0.97 | Hurt |
| Add depth, using 5 layers rather than 3 (ablation) | DGMRF seq 5×5 · Satellite temperatures test set | MAE: 0.95 -> 0.93 | Helped |
| Add depth, using 5 layers rather than 1 (ablation) | DGMRF seq 5×5 · Satellite temperatures test set | MAE: 1.06 -> 0.93 | Helped |
| Use 5×5 rather than 3×3 seq-filters (ablation) | 5-layer DGMRF · Satellite temperatures test set | MAE: 1.16 -> 0.93 | Helped |
| Use 5×5 seq-filters rather than + filters (ablation) | 5-layer DGMRF · Satellite temperatures test set | MAE: 1.09 -> 0.93 | Helped |
| Add nonlinear activations (ablation) | DGMRF seq 5×5, L=5 · Satellite temperatures test set | MAE: 0.93 -> 1.37 | Hurt |
| Integrate linear trend coefficients in the model instead of removing trends by OLS preprocessing (ablation) | DGMRF seq 5×5, L=5 · Satellite temperatures test set | MAE: 1.25 -> 0.93 | Helped |
| Use a linear DGMRF with 3 layers rather than 1 (ablation) | DGMRF · Toy inpainting data with edges | edge preservation | Helped |
Closest papers by meaning
- Scalable Deep Gaussian Markov Random Fields for General Graphs (ICML 2022 · similarity 0.92)
- Deep Gaussian Markov Random Fields for Graph-Structured Dynamical Systems (NeurIPS 2023 · similarity 0.90)
- Scalable Inference of Sparsely-changing Gaussian Markov Random Fields (NeurIPS 2021 · similarity 0.87)
- Deep Random Features for Scalable Interpolation of Spatiotemporal Data (ICLR 2025 · similarity 0.84)
- eXponential FAmily Dynamical Systems (XFADS): Large-scale nonlinear Gaussian state-space modeling (NeurIPS 2024 · similarity 0.83)
- Glauber Generative Model: Discrete Diffusion Models via Binary Classification (ICLR 2025 · similarity 0.83)
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.