Research map / Probabilistic and scientific modeling / Variational and generative sampling
Information criteria for non-normalized models
Takeru Matsuda, Masatoshi Uehara, Aapo Hyvarinen
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 NCIC2 (parameter-count style correction) instead of NCIC1 (tr(IJ^-1) correction) | NCIC bias correction · NCE with Gaussian mixture, N=10^3 | std of bias estimate: around 0.1 -> smaller than 10^{-8} | Improved |
| use SMIC instead of NCIC1 | edge selection · GGM edge (1,2), N=100, sigma12=0.2 | detection probability: 0.515 -> 0.790 | Improved |
| use SMIC instead of AIC | edge selection · GGM edge (1,2), N=100, sigma12=0.2 | detection probability: 0.790 vs 0.783 | No clear effect |
| use NCIC1 instead of NCIC2 | edge selection · GGM edge (1,2), N=100, sigma12=0.2 | detection probability: 0.481 -> 0.515 | Improved |
| increase sigma12 from 0.2 to 0.5 (ablation) | NCIC1 · GGM edge (1,2), N=100 | detection probability: 0.515 -> 0.943 | Improved |
| increase sample size from N=100 to N=200 | NCIC1 · GGM edge (1,2), sigma12=0.2 | detection probability: 0.515 -> 0.749 | Improved |
| use NCIC1 instead of NCE-CV leave-one-out cross-validation | edge selection · Truncated GGM edge (1,2), N=100, sigma12=0.2 | detection probability: 0.210 -> 0.287 | Improved |
| use SMIC instead of SM-CV leave-one-out cross-validation | edge selection · Truncated GGM edge (1,2), N=1000, sigma12=0.2 | detection probability: 0.599 -> 0.623 | No clear effect |
| use log-Gaussian graphical model instead of truncated Gaussian graphical model | SMIC model comparison · RNAseq data (d=40, N=487) | SMIC | Improved |
| add dependence parameter lambda12 to bivariate von Mises model (ablation) | NCIC2 model selection · Wind direction data | NCIC2: -1919 -> -1941 | Improved |
Closest papers by meaning
- Generalized Score Matching for Non-Negative Data (JMLR 2019 · similarity 0.84)
- Conditional Noise-Contrastive Estimation of Unnormalised Models (ICML 2018 · similarity 0.84)
- Statistical Inference with Unnormalized Discrete Models and Localized Homogeneous Divergences (JMLR 2017 · similarity 0.82)
- Scalable high-dimensional Bayesian varying coefficient models with unknown within-subject covariance (JMLR 2023 · similarity 0.81)
- A Unified View on Learning Unnormalized Distributions via Noise-Contrastive Estimation (ICML 2025 · similarity 0.80)
- Kernel Debiased Plug-in Estimation: Simultaneous, Automated Debiasing without Influence Functions for Many Target Parameters (ICML 2024 · similarity 0.80)
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.
