Research map / Probabilistic and scientific modeling / Variational and generative sampling
Gaussian Graphical Model Exploration and Selection in High Dimension Low Sample Size Setting
Thomas Lartigue, Simona Bottani, Stéphanie Baron, Olivier Colliot A, Stanley Durrleman, Stéphanie Allassonnière
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 |
|---|---|---|---|
| replace the GGMselect criterion with cross-validated cross entropy (CVCE) on a fixed graph path | graph selection on synthetic GGM · Synthetic GGM (p=30, n=25/40/100) | true KL divergence: improving the GGMSC choice by a factor from 2 to 5 | Improved |
| use a larger validation fraction (35% or 40%) in CVCE (ablation) | CVCE · Synthetic GGM (p=30) | selected graph closeness to true graph: 35% or 40% of the training data kept in the validation set | Improved |
| use a small validation fraction (20%) in CVCE (ablation) | CVCE · Synthetic GGM (p=30) | variance of performance: 20 % | Hurt |
| add CVCE-based exploration from GGMselect initialisation on top of GGMselect | GGMselect · ADNI (p=343, n=92) | out of sample likelihood: 281 -> 589 or 813 edges | Improved |
| use GLASSO penalised likelihood instead of GGMselect | GGMselect · ADNI (p=343, n=92) | out of sample likelihood: ~3500 edges vs 281 edges | Hurt |
| refit MLE from GLASSO graphs without the l1 penalty (ablation) | GLASSO · Synthetic GGM (p=30 and p=50, n=30) | true cross entropy | No clear effect |
| use Composite method instead of grid GLASSO at small dimension | Composite · Synthetic GGM runtime (p=30) | execution time: 14.9 -> 3.09 | Improved |
| use Composite method instead of grid GLASSO at large dimension | Composite · Synthetic GGM runtime (p=300) | execution time: 1470 -> 6847 | Hurt |
| apply Composite method when p<n (nephrology data) | GGMselect · Nephrology cohort (p=35, n=172) | qualitative graph interpretation | No clear effect |
| vary the GGMselect hyper-parameter K away from 2.5 (ablation) | GGMselect · GGMselect criterion | selected graph | No clear effect |
Closest papers by meaning
- Penalized Maximum Likelihood Estimation of Multi-layered Gaussian Graphical Models (JMLR 2016 · similarity 0.89)
- Evidence Estimation in Gaussian Graphical Models Using a Telescoping Block Decomposition of the Precision Matrix (JMLR 2024 · similarity 0.88)
- Bayesian Covariate-Dependent Gaussian Graphical Models with Varying Structure (JMLR 2022 · similarity 0.88)
- A Generic Family of Graphical Models: Diversity, Efficiency, and Heterogeneity (ICML 2025 · similarity 0.86)
- Learning Local Neighborhoods of Non-Gaussian Graphical Models (AAAI 2025 · similarity 0.85)
- Learning Gaussian DAG Models without Condition Number Bounds (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.
