Logo: to the web site of Uppsala University

In future 6G networks, dependable networks will enable telecommunication services such as remote control of robots or vehicles with strict requirements on end-to-end network performance in terms of delay, delay variation, tail distributions, and throughput. With respect to such networks, it is paramount to be able to determine what performance level the network segment can guarantee at a given point in time. One promising approach is to use predictive models trained using machine learning (ML). Predicting performance metrics such as one-way delay (OWD), in a timely manner, provides valuable insights for the network, user equipments (UEs), and applications to address performance trends, deviations, and violations. Over the course of time, a dynamic network environment results in distributional shifts, which causes catastrophic forgetting and drop of ML model performance. In continual learning (CL), the model aims to achieve a balance between stability and plasticity, enabling new information to be learned while preserving previously learned knowledge. In this paper, we target on the challenges of catastrophic forgetting of OWD prediction model. We propose a novel approach which introducing the concept of multi-generator for the state-of-the-art CL generative replay framework, along with tabular variational autoencoders (TVAE) as generators. The domain knowledge of UE capabilities is incorporated into the learning process for determining generator setup and relevance. The proposed approach is evaluated across a diverse set of scenarios with data that is collected in a realistic 5G testbed, demonstrating its outstanding performance in comparison to baselines.

Intrusion Detection Systems (IDS) are crucial for monitoring and managing critical infrastructure; however, their prominence makes them attractive targets for network attacks. Machine Learning (ML) techniques incorporated into IDS have shown promise in detecting and mitigating these threats. Unfortunately, the scarcity of attack samples presents challenges for model training and generalizability, and attackers can easily bypass detection systems by introducing temporal variations in their attacks. This paper develops strategies for detecting network attacks and specifically examines the impact of network configurations on the generalizability of detection models. As an illustrative example, we investigate the performance of Long Short-Term Memory (LSTM) models in capturing temporal changes in network behavior during attacks, and its resilience against distributional changes. Our study considers multiple factors in terms of attack types, variations, and network configurations, including different topologies and numbers of nodes. We provide insights into how these factors impact the generalizability of models trained using knowledge sharing. To support our research, we implemented Blackhole and DIS-flooding attack variations using the Cooja network simulator. Our objective was to generate a large dataset that enables a comprehensive analysis of attack variations across a diverse set of network configurations, focusing on the impact on LSTM-based IDS for IoT networks.

Dynamic resource allocation for microservices is crucial for meeting management objectives like high throughput and low latency. Recent advancements highlight reinforcement learning as a promising technique, with policy adaptation offering potential for coping with dynamic environments. However, traditional policy adaptation methods often suffer from slow adaptation, particularly in rapidly changing scenarios, limiting their effectiveness in achieving timely responses. To address this limitation, we explore a meta-learning technique to train more generalizable and robust policies that adapt more quickly to target environments. Our findings from extensive experimentation in a real testbed reveal that the MAML meta-learning technique outperforms in scenarios with changing management objectives and achieves comparable performance to baselines in handling dynamic infrastructure loads.

When verifying that a communications network fulfills its specified performance, it is critical to note sudden shifts in network behavior as quickly as possible. Change point detection methods can be useful in this endeavor, but classical methods rely on measuring with a fixed measurement period, which can often be suboptimal in terms of measurement costs. In this paper, we extend the existing framework of change point detection with a notion of physical time. Instead of merely deciding when to stop, agents must now also decide at which future time to take the next measurement. Agents must now minimize the necessary number of measurements pre-and post-change, while maintaining a trade-off between post-change delay and false alarm rate. We establish, through this framework, the suboptimality of typical periodic measurements and propose a simple alternative, called crisis mode agents. We show analytically that crisis mode agents significantly outperform periodic measurements schemes. We further verify this in numerical evaluation, both on an array of synthetic change point detection problems as well as on the problem of detecting traffic load changes in a 5G test bed through end-to-end RTT measurements.

When verifying that a communications network fulfills its specified performance, it is critical to note sudden shifts in network behavior as quickly as possible. Change point detection methods can be useful in this endeavor, but classical methods rely on measuring with a fixed measurement period, which can often be suboptimal in terms of measurement costs. In this paper, we extend the existing framework of change point detection with a notion of physical time. Instead of merely deciding when to stop, agents must now also decide at which future time to take the next measurement. Agents must now minimize the necessary number of measurements pre- and post-change, while maintaining a trade-off between post-change delay and false alarm rate. We establish, through this framework, the suboptimality of typical periodic measurements and propose a simple alternative, called crisis mode agents. We show analytically that crisis mode agents significantly outperform periodic measurements schemes. We further verify this in numerical evaluation, both on an array of synthetic change point detection problems as well as on the problem of detecting traffic load changes in a 5G test bed through end-to-end RTT measurements.

Dynamic resource allocation for network services is pivotal for achieving end-to-end management objectives. Previous research has demonstrated that Reinforcement Learning (RL) is a promising approach to resource allocation in networks, allowing to obtain near-optimal control policies for non-trivial system configurations. Current RL approaches however have the drawback that a change in the system or the management objective necessitates expensive retraining of the RL agent. To tackle this challenge, practical solutions including offline retraining, transfer learning, and model-based rollout have been proposed. In this work, we study these methods and present comparative results that shed light on their respective performance and benefits. Our study finds that rollout achieves faster adaptation than transfer learning, yet its effectiveness highly depends on the accuracy of the system model.

Future 6G networks are expected to be AI-native with distributed machine learning functionalities responsible for improving and automating a variety of network- and service-management tasks. To enable a privacy-preserving approach to distributed learning, federated learning (FL) has become prevalent in the communication-and-networking domain. However, for efficient management of the networks, FL needs to be automated requiring minimal hyperparameter tuning. An outstanding challenge towards automation of FL is regarding difficulties in handling overfitting. Existing techniques tackle overfitting via regularization heuristics that rely on hyperparameter tuning and as such presume availability of representative validation data. However, in the dynamic and heterogeneous network environments, this assumption is limiting. Even if existence of validation data can be assumed, hyperparameter tuning comes with added communication and compute overhead cost which grows prohibitively as the federation scales in size. Here, we propose the congruent federated learning (CFL) as a self-regulated method of learning that is robust to overfitting and achieves the robustness without reliance on hyperparameter tuning. CFL employs a self-taught regularization mechanism that refrains local models from overfitting to the local data. This is enabled via introduction of the congruent activation functions as a class of similarity-promoting activation functions that discourage learning local models which differ excessively from the global (federated) model. Across four networking use cases on several tasks, reflecting different profiles of data heterogeneity and limited availability of data, it is shown that CFL greatly reduces overfitting and in nearly all cases improves the performance-a relative gain of about 21% averaged across all use cases.

From a 5G operator's perspective, accurate estimates of key User Equipments (UEs) performance metrics, especially One-Way Delay (OWD), can provide valuable information. These estimates can trigger management tasks such as reconfiguration to prevent violations of Service Level Objectives (SLOs). Moreover, such insights into UE performance can empower applications to adapt their services to end-users in a more effective manner. We use advanced machine learning over data gathered at the base stations to predict OWD from UEs and show that we are able to predict OWD with over a 2x reduction in percentage error compared to the considered baseline. We discover the close coupling between the performance of the OWD model and the type of UE, which poses a model generalization challenge. Addressing this problem, we demonstrate the shortcomings of the commonly used fine-tuning approach and develop a novel method based on domain adversarial neural networks, that can adapt to a target domain without compromising on the performance of the source domain. Our results show that we can adapt our source model to provide OWD prediction performance within 1-4 percentage points of the ideal scenario when the source and the target domains are the same. Also, our work is grounded in empirical experiments conducted within a 5G testbed, using commercially available hardware.

Intrusion Detection Systems (IDS) play a critical role in safeguarding IoT networks, especially in sectors like healthcare, manufacturing, and smart cities where safety is paramount. Machine learning (ML) holds significant promise for training IDS models, leveraging data from past attacks. However, the effectiveness of these models are dependent on the quality and diversity of training data, which is often limited from the perspective of a single network operator.

This paper delves into the challenges of ML-based IDS model generalization across IoT network scenarios with expected distributional shifts in the data. We examine variations in known attack patterns and changes in IoT network configurations, quantifying their impact on model generalizability. These shifts originates from when multiple network operators seek to share knowledge to enhance their respective IDS capabilities, when a new attack variation is launched, or when an operator reconfigure its network. We explore two approaches to address these challenges: namely data sharing and horizontal federated learning for privacy preservation. While data sharing proves effective across scenarios, it relies on mutual trust among network operators. In contrast, federated learning preserves privacy but is less effective, especially when the network topology is the primary driver of distributional shifts in the train and test data.

To facilitate our study, we implemented Blackhole attack variation strategies within the Cooja network simulator. Our objective was to generate a large dataset enabling comprehensive analysis of attack variations across diverse set of network configurations to study the impact on ML-based IDS for IoT networks.

The future 6G network is envisioned to be AI-native, and as such, ML models will be pervasive in support of optimizing performance, reducing energy consumption, and coping with increasing complexity. A key challenge is automating the process of finding optimal model architectures satisfying stringent requirements stemming from varying tasks, available data, resources in the infrastructure, and deployment positions. In this paper, we look into using Neural Architecture Search (NAS) to automatically find the best model architecture for different telecom use cases. We discuss the main challenges for using NAS in telecom from the perspectives of tabular data, limited data availability, and multi-objective search. We propose a modification to existing NAS methods designed for tabular data to adapt them for multi-objective search and evaluate their performance in different network management and performance modeling scenarios, including scenarios where limited data is available during