[Paper Review] Networks with Memory

Martin Rosvall, Alcides Viamontes Esquivel+3 more|arXiv (Cornell University)|May 21, 2013

COVID-19 epidemiological studies1 references5 citations

TL;DR

This paper introduces memory-aware dynamics on networks to better model real-world flows, showing that memory reduces the influence of highly connected nodes and reveals finer, overlapping community structures. By incorporating memory, the model uncovers more accurate travel patterns in air traffic and scientific communication, improving community detection and ranking beyond traditional memoryless approaches.

ABSTRACT

It is a paradigm to capture the spread of information and disease with random flow on networks. However, this conventional approach ignores an important feature of the dynamics: where flow moves to depends on where it comes from. That is, memory matters. We analyzed multi-step pathways from different systems and show that ignoring the effects of memory has profound consequences for community detection and ranking. Specifi-cally, memoryless dynamics on networks understate the effect of communities and exaggerate the effect of highly connected nodes. For example, including memory reveals actual travel patterns in air traffic, and uncovers multi-disciplinary journals and ranking that favors specialized journals in scientific communication. We also analyzed stochastic spreading processes with and without memory. We find that, when people traveling by plane can trans-mit infections to random other people in cities, the memory effects from their travel patterns are lost. Therefore, the effect of accurately modeling travel patterns on disease spread is small. In systems with limited mixing, how-ever, memory slows down the spread. Overall, memory significantly reduces the effect of highly connected nodes, because the effective node connectivity is lower and the return flows often are many times higher when memory is taken into account. As a result, by only using more available data and making no additional assumptions, we can reveal system organizations with more and smaller modules that overlap to a greater extent. These findings suggest that memory has profound implications for understanding how real systems are organized and function.

Motivation & Objective

• To address the limitation of memoryless network models that fail to capture how flow destinations depend on origins.
• To investigate how memory affects community detection and node ranking in real-world networks.
• To compare memory-aware versus memoryless stochastic spreading processes in systems like air traffic and scientific communication.
• To reveal how memory reduces the dominance of highly connected nodes and exposes overlapping, modular network organizations.

Proposed method

• Analyzing multi-step pathways across diverse systems to quantify memory effects on network dynamics.
• Modeling network flows with memory by tracking the origin of flows, rather than assuming memoryless transitions.
• Comparing stochastic spreading processes with and without memory to assess impacts on disease spread and information diffusion.
• Using real data from air traffic and scientific publication networks to validate memory effects on community structure and node centrality.
• Applying memory-aware dynamics to reveal effective connectivity and return flow patterns that are overlooked in memoryless models.
• Reconstructing network organization using only available data and memory constraints, without additional assumptions.

Experimental results

Research questions

• RQ1 How does incorporating memory in network dynamics alter the perceived importance of highly connected nodes?
• RQ2 To what extent does memory improve the detection of overlapping and finer-grained community structures in real networks?
• RQ3 How do memory effects influence the spread of infections in systems with limited mixing versus well-mixed populations?
• RQ4 In what ways does memory reshape node ranking in scientific communication networks?
• RQ5 Why is the impact of accurate travel pattern modeling on disease spread smaller than expected in systems with memory?

Key findings

• Memory significantly reduces the effective connectivity of highly connected nodes due to increased return flows, diminishing their dominance in network dynamics.
• Memory-aware models reveal more and smaller overlapping communities, indicating a more nuanced network organization than memoryless models.
• Including memory uncovers actual travel patterns in air traffic networks, which are obscured in memoryless models.
• In scientific communication, memory reveals multi-disciplinary journals and favors specialized journals in ranking, challenging traditional centrality measures.
• In systems with limited mixing, memory slows down the spread of infections, whereas in well-mixed systems, the effect of accurate travel patterns is minimal.
• The effective node connectivity is lower under memory-aware dynamics, as repeated return flows reduce the net spread of information or disease.