Introduction to the Special Section on Learning-Based Modeling, Management, and Control for Computer and Communication Networks

Jian Tang, Rong Zheng, Özgür B. Akan, Weiyi ZhangIEEE Trans. Network Science & Engineering 2020 · OpenReview

Abstract

As computer and communication networks grow in scale and complexity, they generate vast quantities of runtime statistics that challenge traditional rule-based management systems. This article introduces a special section of IEEE Transactions on Network Science and Engineering dedicated to leveraging emerging machine learning (ML) techniques—including Deep Neural Networks (DNNs), Deep Reinforcement Learning (DRL), and active learning—to optimize network modeling and control. By synthesizing research across 5G communications, data center management, and social-aware caching, this overview highlights how data-driven approaches are redefining the state-of-the-art in network infrastructure.

Introduction

Modern network infrastructures have reached a level of complexity where manual configuration and static heuristics are no longer sufficient. Every second, these systems produce a staggering volume of data, ranging from traffic loads and resource utilization to signal metrics. While machine learning has already revolutionized fields such as computer vision and natural language processing, its application to the "plumbing" of the internet—the underlying communication networks—is a more recent but equally transformative frontier.

This special section explores how pervasive data can be harnessed to improve network efficiency, reliability, and cost-effectiveness. The featured research demonstrates that ML is not merely a tool for optimization but a fundamental shift in how we approach network architecture.

Methodological Frameworks in Modern Networking

The papers curated for this section employ a diverse array of machine learning methodologies tailored to specific networking challenges:

1. Deep Reinforcement Learning (DRL) for Resource Management

In the context of geo-distributed data centers, managing energy costs is complicated by the unpredictable nature of renewable energy sources. Research in this section utilizes DRL to minimize operational costs by dynamically adjusting to volatile energy capacities, demonstrating the power of agent-based learning in fluctuating environments.

2. Deep Neural Networks (DNNs) for 5G and Optimization

DNNs are applied to two critical areas: predicting Channel State Information (CSI) in 5G systems and identifying "critical links" in wireless networks. By identifying these links, researchers can reduce a massive optimization problem into a smaller, more manageable subset that retains the same optimal solution, significantly lowering computational overhead.

3. Active Learning and Channel Hopping

For activity recognition, a novel approach combines active learning with channel hopping. By selecting high-quality WiFi channels and "hopping" between them, the system creates a virtual wideband channel, providing more robust data than traditional single-channel methods.

4. Hierarchical Clustering for Spectrum Access

To address dynamic spectrum access, a hierarchical framework was developed. This allows nodes with poor connectivity to utilize shared spectrum maps by communicating through "cluster heads," ensuring that even isolated nodes benefit from optimized frequency allocation.

Data-Driven Insights and Social Awareness

A unique aspect of this special section is the recognition that network performance is inextricably linked to human behavior. One study analyzed Twitter data from the 2016 U.S. Presidential Election to develop a context-aware data caching model. By predicting surges in information demand based on social media trends, the network can "pre-cache" content, drastically reducing latency during high-traffic events.

Furthermore, the use of historical CSI data for 5G systems allows for predictive rather than reactive management. Instead of responding to signal degradation after it occurs, these models anticipate changes, allowing for seamless transitions and higher throughput.

Key Results and Performance Gains

The integration of ML into network management has yielded several quantifiable improvements:

  • Latency Reduction: Social-aware caching effectively mitigated delays during peak social events like the 2016 election.
  • Operational Efficiency: Online CSI prediction schemes for 5G proved computationally efficient enough for real-time deployment.
  • Cost Savings: DRL-based models successfully navigated the complexities of green energy markets to reduce data center expenditures.
  • Robustness: The wideband channel-hopping technique improved the accuracy of activity recognition across varied environments compared to standard methods.

Conclusion and Future Directions

The research presented in this special section underscores a pivotal moment in network science. The transition from rule-based systems to learning-based modeling is essential for managing the next generation of communication infrastructure. While these studies provide a robust foundation, the next challenge lies in the large-scale deployment of these models within live, heterogeneous networks. As machine learning becomes a core component of network architecture, the focus will shift toward ensuring these models are as resilient and adaptable as the physical hardware they control.

Summary updated

Abstract

The papers in this special section examine the market for learning-based modeling, management, and control for computer and communication networks. Computer and communication networks are becoming larger and more complicated, generating a huge amount of runtime statistics data (such as traffic load, resource usages, etc.) every second. Meanwhile, emerging machine learning models and techniques, such as active learning, Deep Neural Networks (DNNs) and Deep Reinforcement Learning (DRL), have been shown to dramatically improve the state-of-the-art of many applications, including video/image processing, natural language processing, game playing, etc. This special issue aims to exploit how these emerging and powerful techniques can be leveraged to grasp the exciting opportunities provided by pervasive availability of voluminous data to model, manage and control computer and communication networks. We appreciate contributions to this special section and the valuable and extensive efforts of the reviewers. The topics of this special section range from modeling, analysis, demonstration of various networks with emerging machine learning techniques.
BibTeX
@article{tang2020introduction,
  author = {Jian Tang and Rong Zheng and Özgür B. Akan and Weiyi Zhang},
  title = {{Introduction to the Special Section on Learning-Based Modeling, Management, and Control for Computer and Communication Networks}},
  journal = {IEEE Trans. Network Science \& Engineering},
  year = {2020},
  url = {https://openreview.net/forum?id=UfBLxHDslQw},
}
Figure