Precision Neural Networks: Joint Graph And Relational Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cavallo, Andrea, Rey, Samuel, Marques, Antonio G., Isufi, Elvin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914264520851456
author Cavallo, Andrea
Rey, Samuel
Marques, Antonio G.
Isufi, Elvin
author_facet Cavallo, Andrea
Rey, Samuel
Marques, Antonio G.
Isufi, Elvin
contents CoVariance Neural Networks (VNNs) perform convolutions on the graph determined by the covariance matrix of the data, which enables expressive and stable covariance-based learning. However, covariance matrices are typically dense, fail to encode conditional independence, and are often precomputed in a task-agnostic way, which may hinder performance. To overcome these limitations, we study Precision Neural Networks (PNNs), i.e., VNNs on the precision matrix - the inverse covariance. The precision matrix naturally encodes statistical independence, often exhibits sparsity, and preserves the covariance spectral structure. To make precision estimation task-aware, we formulate an optimization problem that jointly learns the network parameters and the precision matrix, and solve it via alternating optimization, by sequentially updating the network weights and the precision estimate. We theoretically bound the distance between the estimated and true precision matrices at each iteration, and demonstrate the effectiveness of joint estimation compared to two-step approaches on synthetic and real-world data.
format Preprint
id arxiv_https___arxiv_org_abs_2509_14821
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Precision Neural Networks: Joint Graph And Relational Learning
Cavallo, Andrea
Rey, Samuel
Marques, Antonio G.
Isufi, Elvin
Machine Learning
CoVariance Neural Networks (VNNs) perform convolutions on the graph determined by the covariance matrix of the data, which enables expressive and stable covariance-based learning. However, covariance matrices are typically dense, fail to encode conditional independence, and are often precomputed in a task-agnostic way, which may hinder performance. To overcome these limitations, we study Precision Neural Networks (PNNs), i.e., VNNs on the precision matrix - the inverse covariance. The precision matrix naturally encodes statistical independence, often exhibits sparsity, and preserves the covariance spectral structure. To make precision estimation task-aware, we formulate an optimization problem that jointly learns the network parameters and the precision matrix, and solve it via alternating optimization, by sequentially updating the network weights and the precision estimate. We theoretically bound the distance between the estimated and true precision matrices at each iteration, and demonstrate the effectiveness of joint estimation compared to two-step approaches on synthetic and real-world data.
title Precision Neural Networks: Joint Graph And Relational Learning
topic Machine Learning
url https://arxiv.org/abs/2509.14821