Sampling for network function learning

TL;DR

This paper explores the use of graph sampling techniques for learning network functions in valued graphs, addressing challenges when graphs are large, dynamic, or partially unknown.

Contribution

It introduces a framework for applying graph sampling to network function learning and discusses associated learning methods for incomplete or large-scale graphs.

Findings

Sampling enables learning in large or dynamic graphs

Methods can handle incomplete edge information

Sampling maintains key network function properties

Abstract

Given a valued graph, where both the nodes and the edges of the graph are associated with one or several values, any network function for a given node must be defined in terms of that node and its connected nodes in the graph. Generally, applying the same definition to the whole graph or any given subgraph of it would result in systematically different network functions. In this paper we consider the feasibility of graph sampling approach to network function learning, as well as the corresponding learning methods based on the sample graphs. This can be useful either when the edges are unknown to start with or the graph is too large (or dynamic) to be processed entirely.