Research map / Learning theory and algorithms / Matrix and tensor methods
Graphical Lasso and Thresholding: Equivalence and Closed-form Solutions
Salar Fattahi, Somayeh Sojoudi
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 |
|---|---|---|---|
| use closed-form solution instead of QUIC | graphical lasso · Large-scale GL d=20000, m=99954 | runtime (s): 1266.5 -> 39.4 | Improved |
| use closed-form solution instead of GLASSO with warm start | graphical lasso · Large-scale GL d=16000, m=160094 | runtime (s): 9960.8 -> 18.5 | Improved |
| use closed-form solution instead of elementary estimator | graphical lasso · Large-scale GL d=20000, m=200018 | runtime (s): 211.5 -> 37.4 | Improved |
| use closed-form solution instead of QUIC and GLASSO at d=40000 | graphical lasso · Large-scale GL d=40000, m=200290 | solvable within 4h limit: 495.4 | Improved |
| use closed-form solution at d=80000 (3.2 billion variables) | graphical lasso · Large-scale GL d=80000, m=401798 | runtime (s): 1450.4 | Improved |
| add warm-start (thresholding-based) initialization to QUIC | QUIC · Large-scale GL d=20000, m=99954 | runtime (s): 1266.5 -> 539.4 | Improved |
| add warm-start initialization to GLASSO | GLASSO · Large-scale GL d=8000, m=40218 | runtime (s): 3675.1 -> 1011.2 | Improved |
| add warm-start initialization to GLASSO enabling larger instances | GLASSO · Large-scale GL d=12000-16000 | solvable instance size | Improved |
| increase lambda on brain network covariance (ablation) | closed-form approximation · Brain network (140 regions) | edge mismatches vs optimal: less than 10% above 0.35, almost 0 above 0.5 | Improved |
| increase lambda on brain network covariance (ablation) | closed-form approximation · Brain network (140 regions) | 2-norm approximation error / similarity degree: error less than 0.37, similarity greater than 0.98 when lambda above 0.4 | Improved |
| increase lambda on traffic network covariance (ablation) | closed-form approximation · Traffic network (1049 stations) | positive-definiteness and mismatches: PD above 0.991, mismatches almost zero above 0.97 | Improved |
| increase cycle length in the sparsity graph (ablation) | closed-form solution · Synthetic cycle graph | optimality gap: less than 10^-6 when cycle length at least 6 | Improved |
Closest papers by meaning
- Equivalence of Graphical Lasso and Thresholding for Sparse Graphs (JMLR 2016 · similarity 0.91)
- Large-Scale Sparse Inverse Covariance Estimation via Thresholding and Max-Det Matrix Completion (ICML 2018 · similarity 0.90)
- Testing for Differences in Gaussian Graphical Models: Applications to Brain Connectivity (NeurIPS 2016 · similarity 0.88)
- Joint Structural Estimation of Multiple Graphical Models (JMLR 2016 · similarity 0.88)
- Tensor Graphical Model: Non-Convex Optimization and Statistical Inference (TPAMI 2019 · similarity 0.88)
- High-Dimensional Inference for Cluster-Based Graphical Models (JMLR 2020 · similarity 0.87)
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.
