Machine Learning as a Network Management Primitive - From End-to-End Optimization to Atomic Network Functions

Aug 3, 2026

Upload by Skyler

Machine Learning (ML) is positioned as a data-driven foundation for network management, enabling many task-specific functions within the networking stack. Current research has focused on training accurate models from historical data, yet the full lifecycle remains incomplete: deploying, operating, and maintaining ML solutions introduces orthogonal research and engineering challenges. This dissertation addresses five core challenges—generalizability, adaptability, reliability guarantees, data efficiency, and measurable performance—through multiple application-driven studies and extensive validation on novel real-world datasets released publicly.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages152

Reading Time6 h 23 min

AccessFree to download

Abstract

Contents

Introduction

Problem Statement

Thesis Structure

Other Contributions

State of the Art

Machine Learning for Network Optimization

Machine Learning for Failure Identification

Machine Learning for Service Orchestration

In-Network Machine Learning

Language Models for Network Automation

Reinforcement Learning for Network Optimization

On Reinforcement Learning for Routing and Wavelength Assignment

Reinforcement-Learning-based Local Search for Network Optimization

Data-Centric and Reliable Machine Learning for Fault Management

Data-Centric Machine Learning for Hardware Fault Classification

As-Soon-As-Possible Hardware Fault Classification with Guarantees

Machine Learning for Service Orchestration

Scalable Service Orchestration with Reinforcement Learning

Reinforcement Learning for Multi-Objective Service Orchestration

Continual In-Network Learning

Large Language Models for Automated Network Configuration

Conclusion and future developments

Conclusion

Future developments

Machine Learning as a Network Management Primitive: From End-to-End Optimization to Atomic Network Functions Doctoral Dissertation of: Nicola Di Cicco Advisor: Prof. Massimo Tornatore Tutor: Prof. Ilario Filippini Year 2023/2024-XXXVII Cycle Abstract Machine Learning (ML) is rapidly becoming the jack-of-all-trades of the network management stack. Thanks to its purely data-driven nature, ML can be leveraged for developing a broad set of network management functions tailored for the specific task at hand. Unfortunately, we are still far from harnessing the full potential of ML for network management. Contemporary literature in this field has made tremendous progress in identifying algorithms and design principles for producing well-trained models from historical data. These contributions, though fundamental, cover only one aspect of the whole ML model lifecycle. Deploying, using, and maintaining ML-based network management solutions poses nontrivial research challenges, orthogonal with respect to training a model, that is presently not yet thoroughly addressed. This hinders a widespread adoption of ML by network operators, who instead prefer relying upon classical, battletested network management solutions. To take a step toward solving this fundamental problem, this Thesis identifies five core challenges in ML for network management: 1) Generalizability to data beyond training, 2) Adaptability to dynamic network environments, 3) Reliability, in terms of providing theoretical performance guarantees after model deployment, 4) Data efficiency, for minimizing the amount of labor required for training and updating models, and 5) Performance, i.e., ML-based solutions must tangibly improve over conventional methods to be worth considering. We address these challenges through multiple representative network management applications: online and offline network optimization with Reinforcement Learning, focusing on generalizability and performance; ML-based hardware fault classification in microwave networks, focusing on data efficiency and reliability; Continual in-network ML, focusing on data efficiency and adaptability; and intent-based networking with Large Language Models, focusing on performance and data-efficiency. We quantitatively validate the practical effectiveness of our proposed solutions through extensive comparisons against the state of the art, and by leveraging novel real-world datasets, which we make publicly available. Keywords: Machine Learning; Network Management; Optimization iii Contents Abstract i Contents iii 1 Introduction 1 1.1 Problem Statement ............................... 2 1.2 Thesis Structure ................................. 5 1.3 Other Contributions .............................. 14 2 State of the Art 19 2.1 Machine Learning for Network Optimization ................. 19 2.2 Machine Learning for Failure Identification .................. 22 2.3 Machine Learning for Service Orchestration ................. 26 2.4 In-Network Machine Learning ......................... 28 2.5 Language Models for Network Automation .................. 29 3 Reinforcement Learning for Network Optimization 31 3.1 On Reinforcement Learning for Routing and Wavelength Assignment ... 31 3.2 Reinforcement-Learning-based Local Search for Network Optimization ... 45 4 Data-Centric and Reliable Machine Learning for Fault Management 59 4.1 Data-Centric Machine Learning for Hardware Fault Classification ..... 59 4.2 As-Soon-As-Possible Hardware Fault Classification with Guarantees .... 73 5 Machine Learning for Service Orchestration 85 5.1 Scalable Service Orchestration with Reinforcement Learning ........ 85 5.2 Reinforcement Learning for Multi-Objective Service Orchestration ..... 96 6 Continual In-Network Learning 109 7 Large Language Models for Automated Network Configuration 113 8 Conclusion and future developments 119 8.1 Conclusion .................................

FAQs

What problem does the thesis target in ML for network management?

It targets the gap between training well-performing ML models and the additional, nontrivial challenges of deploying, using, and maintaining ML-based network management solutions in practice.

Which five core challenges are identified for ML in network management?

The thesis identifies generalizability beyond training data, adaptability to dynamic environments, reliability with theoretical performance guarantees after deployment, data efficiency to reduce training/update labor, and performance that must improve over conventional methods.

How does the thesis validate its proposed solutions?

It uses extensive quantitative comparisons against state of the art and leverages novel real-world datasets that are made publicly available.

Aug 3, 2026

Upload by Skyler