Machine Learning for Networking: Workflow, Advances and Opportunities (1709.08339v2)

Abstract: Recently, machine learning has been used in every possible field to leverage its amazing power. For a long time, the net-working and distributed computing system is the key infrastructure to provide efficient computational resource for machine learning. Networking itself can also benefit from this promising technology. This article focuses on the application of Machine Learning techniques for Networking (MLN), which can not only help solve the intractable old network questions but also stimulate new network applications. In this article, we summarize the basic workflow to explain how to apply the machine learning technology in the networking domain. Then we provide a selective survey of the latest representative advances with explanations on their design principles and benefits. These advances are divided into several network design objectives and the detailed information of how they perform in each step of MLN workflow is presented. Finally, we shed light on the new opportunities on networking design and community building of this new inter-discipline. Our goal is to provide a broad research guideline on networking with machine learning to help and motivate researchers to develop innovative algorithms, standards and frameworks.

• The paper introduces a structured MLN workflow from problem formulation to deployment, enhancing network problem-solving.
• It surveys ML techniques like hidden Markov models and deep reinforcement learning for traffic prediction and resource management.
• The paper highlights challenges in real-time deployment, urging further research in data robustness and adaptive scheduling.

Machine Learning for Networking: Workflow, Advances, and Opportunities

The intersection of ML and networking, referred to as Machine Learning for Networking (MLN), provides a compelling domain for research that intersects robust computational algorithms with complex network system environments. This paper, authored by Wang et al., explores the MLN framework, offering a holistic view of how machine learning techniques can enhance various facets of networking.

Overview of MLN Workflow

The framework for MLN is built upon a structured workflow composed of specific stages: Problem Formulation, Data Collection, Data Analysis, Model Construction, Model Validation, and Deployment and Inference. Each stage builds upon the disciplines of traditional machine learning and adapts them to the unique requirements of network applications, allowing for the innovative use of ML capabilities to address network challenges.

1. Problem Formulation: Here, each networking task is abstracted appropriately to fit into a machine learning paradigm, such as classification, clustering, or decision-making.
2. Data Collection: Emphasis is placed on acquiring a robust dataset, encompassing both historical data and real-time network state data, which forms the foundation for modeling.
3. Data Analysis: This phase involves preprocessing and feature extraction tailored to the network problem at hand, serving as a crucial determinant of the model's effectiveness.
4. Model Construction: Utilizes suitable algorithms contingent on the problem and dataset size, incorporating hyper-parameter tuning.
5. Model Validation: Engages in cross-validation to ensure that the model effectively predicts, thereby aiding in fine-tuning the model for optimal performance.
6. Deployment and Inference: Concerns the practical integration of ML models in live environments, addressing issues like computational limits and inference speed.

Evaluation of Advances

The paper presents a survey of cutting-edge research in MLN, categorizing recent studies across diverse networking applications such as traffic prediction, traffic classification, TCP congestion control, and resource management.

• Traffic Prediction and Classification: Techniques such as Hidden-Markov Models and clustering approaches are applied to anticipate network performance and classify traffic types, delivering improved predictive capabilities and supporting network security.
• Resource Management and Network Adaptation: Here, reinforcement learning and deep learning techniques are deployed to address decision-making challenges in environments where traditional heuristic algorithms fall short. The deployment of deep reinforcement learning is noted for its efficacy in resource scheduling tasks.
• Network Performance Prediction and Configuration Extrapolation: Approaches using models to predict system states allow optimization in video QoE and network throughput, transcending traditional statistical methodologies by harnessing the predictive precision of ML techniques.

Implications and Opportunities

The paper positions MLN as a burgeoning interdisciplinary field with significant prospects. However, the paper succinctly notes challenges, such as model feasibility in real-time systems, robust data acquisition, and accountability of ML models in deployment scenarios.

Future research is poised to address these challenges by expanding open datasets, improving network protocol and architecture design, and enhancing ML model robustness and generalization. The work suggests that automated protocol design and adaptive scheduling will become paramount as network environments grow increasingly dynamic and complex.

In conclusion, this paper provides a pertinent overview of how machine learning can revolutionize the networking domain. It serves as both a comprehensive guide for practitioners in ML and networking and as a springboard for future research to navigate an evolving technological landscape.