Research map / Probabilistic and scientific modeling / Variational and generative sampling
Generalized Matrix Factorization: efficient algorithms for fitting generalized linear latent variable models to large data arrays
Lukasz Kidzinski, Francis K.C. Hui, David I. Warton, Trevor J. Hastie
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 |
|---|---|---|---|
| add latent variables (p=3) to the fixed-effects-only model (ablation) | Newton GLLVM fit · Plant species (48331x2211, Bernoulli, held-out 1000 entries) | AUC: 0.72 -> 0.87 | Improved |
| add latent variables (p=3) to the fixed-effects-only model (ablation) | Newton GLLVM fit · Plant species (48331x2211, Bernoulli, held-out 1000 entries) | out-of-sample deviance explained: 39% -> 58% | Improved |
| use PQL with Newton method instead of variational approximation (gllvm) | GLLVM fitting · Ant abundance (41x30, Poisson) | computation time: 2.3 seconds -> 0.2 seconds | Improved |
| run AIRWLS in parallel across 8 CPUs instead of 1 CPU | AIRWLS · Ant abundance (41x30, Poisson) | computation time: 0.5 seconds -> 0.04 seconds | Improved |
| use proposed Newton instead of gllvm | GLLVM fitting · Simulations (aggregate over n, m, p) | mean computation time: 51.6 minutes -> 51 seconds | Improved |
| use proposed methods instead of gllvm when sample size is very small (n=5) | GLLVM fitting · Simulations (small sample, m=20, n=5) | deviance explained: 0.922 -> 0.733 (AIRWLS), 0.65 (Newton) | Hurt |
| use AIRWLS instead of gllvm at large latent dimension (p=10, n=500, m=300) | GLLVM fitting · Simulations (p=10, n=500, m=300) | computation time: 2.5 days -> 40 seconds | Improved |
| switch response distribution from Poisson to binomial | gllvm, AIRWLS, Newton · Simulations | mean deviance explained: 0.383, 0.434, 0.412 -> 0.092, 0.127, 0.125 | Hurt |
| choose latent dimension by smooth regularization (gamma) instead of rank constraint (p) | Newton GLLVM fit · Simulations (Poisson, n=m=100, p=2, model selection) | held-out mean deviance: 1.279 -> 1.272 | No clear effect |
| drop a latent variable as covariate proxy: removed Shrub.cover and checked second latent variable | AIRWLS GLLVM · Ant abundance (41x30, Poisson) | Pearson correlation with held-out Shrub.cover: -0.49 | Improved |
Closest papers by meaning
- Bayesian group factor analysis with structured sparsity (JMLR 2016 · similarity 0.87)
- skglm: Improving scikit-learn for Regularized Generalized Linear Models (JMLR 2025 · similarity 0.87)
- Graphical Lasso and Thresholding: Equivalence and Closed-form Solutions (JMLR 2019 · similarity 0.87)
- Speeding Up Latent Variable Gaussian Graphical Model Estimation via Nonconvex Optimization (NeurIPS 2017 · similarity 0.87)
- Algebraic tests of general Gaussian latent tree models (NeurIPS 2018 · similarity 0.86)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.84)
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.
