Paper Title:
Estimation of partially known Gaussian graphical models with score-based structural priors
Published on:
25 January 2024
Primary Category:
Machine Learning
Paper Authors:
Martín Sevilla,
Antonio García Marques,
Santiago Segarra
Proposes novel way to estimate Gaussian graphical models using partial graph data
Leverages graph neural networks to learn prior over graph structures
Uses Langevin dynamics to sample from posterior over unknown graph edges
Demonstrates improved performance over existing methods in experiments
Estimating Gaussian graphical models from partial graphs
This paper proposes a new method to estimate the structure of Gaussian graphical models when part of the graph is already known. It uses graph neural networks and a sampling technique called Langevin dynamics to incorporate flexible prior information about graph structure from existing graph datasets. This allows it to outperform classical approaches in experiments.
Neural networks learn to sample from distributions by approximating Langevin dynamics
Learning networks from Gaussian graphical models
Learning dependence structure from data
Scalable Bayesian neural networks via function-space variational inference
Edge flows on networks
Geometry-Aware 3D Gaussian Modeling for Dynamic Scenes
No comments yet, be the first to start the conversation...
Sign up to comment on this paper