Reinforcement Learning (RL) has emerged as a powerful tool for automating complex network management tasks, yet its lack of transparency and black-box nature hinder trust and adoption in operational environments. In this work, we focus on explaining the behavior of an $\mathbf{R L}$ agent applied to the problem of network slice admission control. We present a framework that integrates three key components: a Deep Reinforcement Learning (DRL) agent for admission control, an Integer Linear Programming (ILP) model for network slice embedding, and an explanation module for interpreting the DRL agent’s policies, namely Shapley Value Explainable Reinforcement Learning (SVERL). Our analysis aims gives particular attention to cases where the RL agent rejects admitting a network slice request despite sufficient network capacity to provision it, and investigates whether explanations can be used to verify and validate the agent’s behavior prior to deployment approval. Experimental results reveal that the agent’s decisions are primarily influenced by substrate network conditions such as congestion, rather than by the intrinsic characteristics of slice requests. While this conservative policy prevents overload, it also leads to overly cautious rejections. Importantly, the proposed explanation framework provides operators with actionable insights to scrutinize, validate, and refine RL-driven policies before operational deployment.

Integrating advanced Machine Learning (ML) techniques, such as Federated Learning (FL), with emerging paradigms like Multi-access Edge Computing (MEC) and Software-Defined Wide Area Network (SD-WAN) has gained significant attention. In this demo, we present a simulator designed to optimize resource allocation for FL in MEC SD-WAN environments, taking into account network capacity constraints. We demonstrate in real-time how resource allocation is implemented for FL algorithms and visually analyze the impact of different constraints on performance. Additionally, we introduce a heuristic algorithm for Distributed Federated Learning (DFL) that selects the optimal global aggregator based on resource utilization. We evaluate the proposed heuristic within our simulation framework and compare its performance against Centralized Federated Learning (CFL) and traditional ML techniques. Our results demonstrate how the proposed simulator effectively integrates network and resource allocation, providing valuable insights into the interplay between FL and MEC SD-WAN infrastructure.

Software-Defined Wide Area Networks (SD-WANs) have emerged as a transformative solution for modern enterprise networking, enabling dynamic traffic management, cost-efficient connectivity, and improved network performance. However, ensuring real-time visibility into network conditions remains a key challenge, as SD-WAN overlay tunnels operate over diverse and often unpredictable underlay networks. Traditional network monitoring techniques, such as active and passive monitoring, face limitations in balancing accuracy, responsiveness, and overhead. To address this challenge, we propose an In-Band Network Telemetry (INT) framework for SD-WANs, leveraging extended Berkeley Packet Filter (eBPF) technology for efficient and flexible packet processing. Our approach enables real-time telemetry data collection at the Customer Premises Equipment (CPE) level, allowing for precise performance monitoring while minimizing additional network overhead. The framework integrates seamlessly with various VPN-based SD-WAN tunnels, including Generic Routing Encapsulation (GRE), IP Security (IPSec), and IPSec over GRE, ensuring adaptability across different deployment scenarios. By embedding telemetry metadata directly into overlay packets, the proposed solution provides continuous monitoring of critical Quality of Service (QoS) metrics, such as One-Way Delay (OWD), Two-Way Delay (TWD), and packet loss rate. Through extensive experimentation, we demonstrate the effectiveness of our INT-enabled SD-WAN framework in accurately detecting network anomalies and ensuring Service-Level Agreement (SLA) compliance. The results validate our approach as a scalable and lightweight monitoring solution for enhancing network observability in SD-WAN deployments.

Wide Area Networks (WANs) are essential for enabling communication among different branches within enterprises, ultimately improving operational efficiency and fostering economic advancement. However, the traditional WAN solutions often struggles to meet the diverse needs of geographically dispersed organizations. The advent of Software-Defined Wide Area Network (SD-WAN) solutions represents a paradigm shift, harnessing Software-Defined Networking (SDN) principles to transcend the limitations inherent in traditional WAN architectures. However, the necessity for advanced monitoring solutions persists as fundamental in guaranteeing optimal performance and unwavering reliability. This paper proposes the integration of In-band Network Telemetry in SD-WANs through Extended Berkeley Packet Filter (eBPF) technology to assess network performance parameters, such as the delay, in real-time. This work addresses the increasing demand for performance, security and reliability in such network architectures by leveraging INT capabilities for monitoring and manipulating data packets. Through the exploitation of eBPF’s features and network programmability, we introduce a novel approach aimed at measuring the end-to-end delay across SD-WAN deployments.

Wild animals pose significant challenges in agricultural environments, necessitating effective monitoring solutions for both agricultural management and wildlife preservation. Leveraging novel deep learning algorithms for real-time animal detection holds great promise in enhancing wildlife monitoring activities. This work presents a performance evaluation of two deep-learning based image recognition models (YOLOv8 and YOLOv9) with application of color space augmentation to simulate different lightning scenarios in real-life environments. Our results show that the detection accuracy is increased by the augmented color spaces versus the natural ones, as the proposed technique is able to strengthen the model’s robustness against environmental changes. Furthermore, we consider the deployment of these models on far edge devices, such as trap cameras with GPUs, where real-time analysis of wildlife activity is crucial for both management and conservation efforts.