Generalised Score Matching on Convex Domains
Abstract
Score matching avoids computing the normalising constant that maximum-likelihood estimation requires. On constrained domains, its generalised variants weight the Fisher divergence so that boundary terms vanish. We derive generalised score matching on open convex subsets of ℝd as the small-neighbourhood limit of minimum probability flow, in which the geometry of the neighbourhoods determines the weight. Every C2 positive definite weight arises in this way, including those of classical score matching on ℝd and of its variants for non-negative data on ℝ+d. For exponential families, we extend the standard convexity, consistency and asymptotic normality results to every such weight and show that the estimator converges to the true parameter under certain boundary conditions. For a truncated Gaussian on a polytope and a Dirichlet distribution on the simplex, proposed estimators attain the lowest median error of all methods compared, in at least 42 of 50 ground-truth configurations.
Related
- Weighted Score Matching on Bounded Domains: Nonparametric Density Estimation and Interaction Graph Recovery
- SLOPE DETERMINATION ON CONVEX DOMAINS: DIRICHLET DATA AND CRANDALL-PAZY DIRECTIONS
- Kobayashi visibility of strictly convex domains
- Strongly linearly convex exhaustion of a class of ℂ-convex domains
- The Table Problem for Convex Domains
BibTeX
@misc{shetty2026generalised,
title = {{Generalised Score Matching on Convex Domains}},
author = {Nishanth Shetty and Saisuchith Mahajan and Chandra Sekhar Seelamantula},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2609.11521},
eprint = {2609.11521},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.11521},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Generalised Score Matching on Convex Domains
AU - Nishanth Shetty
AU - Saisuchith Mahajan
AU - Chandra Sekhar Seelamantula
PY - 2026
DA - 2026/10/07
PB - arXiv
DO - 10.48550/arxiv.2609.11521
UR - https://arxiv.org/abs/2609.11521
AB - Score matching avoids computing the normalising constant that maximum-likelihood estimation requires. On constrained domains, its generalised variants weight the Fisher divergence so that boundary terms vanish. We derive generalised score matching on open convex subsets of ℝ^d as the small-neighbourhood limit of minimum probability flow, in which the geometry of the neighbourhoods determines the weight. Every C^2 positive definite weight arises in this way, including those of classical score matching on ℝ^d and of its variants for non-negative data on ℝ_+^d. For exponential families, we extend the standard convexity, consistency and asymptotic normality results to every such weight and show that the estimator converges to the true parameter under certain boundary conditions. For a truncated Gaussian on a polytope and a Dirichlet distribution on the simplex, proposed estimators attain the lowest median error of all methods compared, in at least 42 of 50 ground-truth configurations.
ER -