Embedding-aided network dismantling

TL;DR

This paper introduces an embedding-based approach for network dismantling that leverages geometric representations, providing a unified and effective method applicable to various network types and variants.

Contribution

It proposes a novel embedding-aided strategy for network dismantling that works across different geometric embeddings and outperforms existing algorithms in many cases.

Findings

Embedding-based methods are effective for network dismantling.

The approach works well on both synthetic and real networks.

Performance is comparable or superior to current state-of-the-art algorithms.

Abstract

Optimal percolation concerns the identification of the minimum-cost strategy for the destruction of any extensive connected components in a network. Solutions of such a dismantling problem are important for the design of optimal strategies of disease containment based either on immunization or social distancing. Depending on the specific variant of the problem considered, network dismantling is performed via the removal of nodes or edges, and different cost functions are associated to the removal of these microscopic elements. In this paper, we show that network representations in geometric space can be used to solve several variants of the network dismantling problem in a coherent fashion. Once a network is embedded, dismantling is implemented using intuitive geometric strategies. We demonstrate that the approach well suits both Euclidean and hyperbolic network embeddings. Our systematic analysis on synthetic and real networks demonstrates that the performance of embedding-aided techniques is comparable to, if not better than, the one of the best dismantling algorithms currently available on the market.