Aggregate Graph Statistics

TL;DR

This paper demonstrates how aggregate graph statistics, computed via a distributed HyperANF algorithm mapped onto field calculus, can enhance collective adaptive systems like IoT networks.

Contribution

It introduces a fully distributed, asynchronous implementation of HyperANF using field calculus, enabling scalable graph analysis in dynamic IoT environments.

Findings

HyperANF can be mapped to field calculus for distributed execution

The approach supports asynchronous, massively parallel graph computations

Provides a new self-stabilising building block for aggregate computing

Abstract

Collecting statistic from graph-based data is an increasingly studied topic in the data mining community. We argue that these statistics have great value as well in dynamic IoT contexts: they can support complex computational activities involving distributed coordination and provision of situation recognition. We show that the HyperANF algorithm for calculating the neighbourhood function of vertices of a graph naturally allows for a fully distributed and asynchronous implementation, thanks to a mapping to the field calculus, a distribution model proposed for collective adaptive systems. This mapping gives evidence that the field calculus framework is well-suited to accommodate massively parallel computations over graphs. Furthermore, it provides a new "self-stabilising" building block which can be used in aggregate computing in several contexts, there including improved leader election or network vulnerabilities detection.