Foundations and Modeling of Dynamic Networks Using Dynamic Graph Neural Networks: A Survey

Skarding, Joakim; Gabrys, Bogdan; Musiał, Katarzyna

doi:10.1109/access.2021.3082932

Cited by 204 publications

(130 citation statements)

References 83 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For example, networks may be polarized (Lofdahl et al, 2015 ), hierarchical (Giabbanelli, 2011 ), planar (Giabbanelli, 2010 ), or dynamic. This diversity is exemplified in generators of dynamic networks, which use methods as varied as stochastic blockmodels (Kim et al, 2018 ), graph neural networks (Skardinga et al, 2021 ), or event models (Fritz et al, 2020 ). A potential follow-up study could thus extend our approach by including network properties to assess whether they have pronounced effects on the cost of computing various measures.…”

Section: Discussionmentioning

confidence: 99%

An Experimental Study on the Scalability of Recent Node Centrality Metrics in Sparse Complex Networks

Freund¹,

Giabbanelli²

2022

Front. Big Data

View full text Add to dashboard Cite

Node centrality measures are among the most commonly used analytical techniques for networks. They have long helped analysts to identify “important” nodes that hold power in a social context, where damages could have dire consequences for transportation applications, or who should be a focus for prevention in epidemiology. Given the ubiquity of network data, new measures have been proposed, occasionally motivated by emerging applications or by the ability to interpolate existing measures. Before analysts use these measures and interpret results, the fundamental question is: are these measures likely to complete within the time window allotted to the analysis? In this paper, we comprehensively examine how the time necessary to run 18 new measures (introduced from 2005 to 2020) scales as a function of the number of nodes in the network. Our focus is on giving analysts a simple and practical estimate for sparse networks. As the time consumption depends on the properties in the network, we nuance our analysis by considering whether the network is scale-free, small-world, or random. Our results identify that several metrics run in the order of O(nlogn) and could scale to large networks, whereas others can require O(n2) or O(n3) and may become prime targets in future works for approximation algorithms or distributed implementations.

show abstract

Section: Discussionmentioning

confidence: 99%

An Experimental Study on the Scalability of Recent Node Centrality Metrics in Sparse Complex Networks

Freund¹,

Giabbanelli²

2022

Front. Big Data

View full text Add to dashboard Cite

show abstract

“…Over the last few years, multiple interesting literature surveys and reviews [1][2][3][4][5][6][7][8][9] have been done which shown the length and breadth of this emerging area. This is accompanied by their rich and growing research application in natural sciences, recommender systems, or day-to-day use systems (traffic networks, social networks, computer networking).…”

Section: Introductionmentioning

confidence: 99%

Graph Neural Networks and Application for Cosmic-Ray Analysis

Koundal¹

2021

Proceedings of the 5th International Workshop on Deep Learning in Computational Physics — PoS(DLCP2021)

View full text Add to dashboard Cite

Deep Learning has emerged as one of the most promising areas of computational research for pattern learning, inference drawing, and decision-making, with wide-ranging applications across various scientific disciplines. This has also made it possible for faster and more precise analysis in astroparticle physics, enabling new insights from massive volumes of input data. Graph Neural Networks have materialized as a salient implementation method among the numerous deeplearning architectures over the last few years because of the unique ability to represent complex input data from a wide range of problems in its most natural form. Described using nodes and edges, graphs allow us to efficiently represent relational data and learn hidden representations of input data to obtain better model accuracy. At IceCube Neutrino Observatory, a complex multicomponent detector, traditional likelihood-based analysis on a per-event basis, to reconstruct cosmic-ray air shower parameters is time-consuming and computationally costly. Using advanced and flexible models based on Graph Neural Networks naturally emerges as a possible solution, reducing the time and computing cost for performing such analysis while boosting sensitivity. This paper will outline Graph Neural Networks and discuss a possible application of using such methods at the IceCube Neutrino Observatory.

show abstract

“…Representation learning on temporal interaction networks has gradually become a hot topic in the research of machine learning [24]. A key challenge of modeling temporal interaction networks is how to capture the evolution of user interests and item features effectively.…”

Section: Introductionmentioning

confidence: 99%

Deep Structural Point Process for Learning Temporal Interaction Networks

Cao¹,

Lin²,

Cong³

et al. 2021

Preprint

View full text Add to dashboard Cite

This work investigates the problem of learning temporal interaction networks. A temporal interaction network consists of a series of chronological interactions between users and items. Previous methods tackle this problem by using different variants of recurrent neural networks to model interaction sequences, which fail to consider the structural information of temporal interaction networks and inevitably lead to sub-optimal results. To this end, we propose a novel Deep Structural Point Process termed as DSPP for learning temporal interaction networks. DSPP simultaneously incorporates the topological structure and long-range dependency structure into the intensity function to enhance model expressiveness. To be specific, by using the topological structure as a strong prior, we first design a topological fusion encoder to obtain node embeddings. An attentive shift encoder is then developed to learn the long-range dependency structure between users and items in continuous time. The proposed two modules enable our model to capture the user-item correlation and dynamic influence in temporal interaction networks. DSPP is evaluated on three real-world datasets for both tasks of item prediction and time prediction. Extensive experiments demonstrate that our model achieves consistent and significant improvements over state-of-the-art baselines.

show abstract

Foundations and Modeling of Dynamic Networks Using Dynamic Graph Neural Networks: A Survey

Cited by 204 publications

References 83 publications

An Experimental Study on the Scalability of Recent Node Centrality Metrics in Sparse Complex Networks

An Experimental Study on the Scalability of Recent Node Centrality Metrics in Sparse Complex Networks

Graph Neural Networks and Application for Cosmic-Ray Analysis

Deep Structural Point Process for Learning Temporal Interaction Networks

Contact Info

Product

Resources

About