A comparative analysis of knowledge acquisition performance in complex networks

TL;DR

This study systematically compares knowledge acquisition performance across various network topologies and dynamics, revealing that different strategies can produce similar learning curves, which complicates inferring network structure solely from these curves.

Contribution

It provides a comprehensive analysis of how different network topologies and dynamics influence knowledge acquisition performance, highlighting ambiguities in feature space and implications for network inference.

Findings

All learning curves share the same shape but differ in speed.

Different network topologies can produce similar learning curves.

Learning curves alone are insufficient for inferring network topology.

Abstract

Discovery processes have been an important topic in the network science field. The exploration of nodes can be understood as the knowledge acquisition process taking place in the network, where nodes represent concepts and edges are the semantical relationships between concepts. While some studies have analyzed the performance of the knowledge acquisition process in particular network topologies, here we performed a systematic performance analysis in well-known dynamics and topologies. Several interesting results have been found. Overall, all learning curves displayed the same learning shape, with different speed rates. We also found ambiguities in the feature space describing the learning curves, meaning that the same knowledge acquisition curve can be generated in different combinations of network topology and dynamics. A surprising example of such patterns are the learning curves obtained from random and Waxman networks: despite the very distinct characteristics in terms of global structure, several curves from different models turned out to be similar. All in all, our results suggest that different learning strategies can lead to the same learning performance. From the network reconstruction point of view, however, this means that learning curves of observed sequences should be combined with other sequence features if one aims at inferring network topology from observed sequences.