arXiv subjects

Mounir Bensalem

Publications and source records attributed to Mounir Bensalem.

At least 19 recordsLinked to original sources

arXiv2026

A Self-Calibrating Agentic AI Framework for Autonomous Edge Resource Allocation

Large Language Models (LLMs) are increasingly deployed as autonomous agents, transitioning from static conversational interfaces to dynamic systems capable of complex reasoning, tool execution, and decision-making. However, the operational reliability of these agentic AI systems is fundamentally challenged by the absence of reliable ground truth in open-ended environments and the risk of increasing operational drift over time. To address this challenge, we propose and experimentally evaluate an agentic AI framework, designed to enforce autonomous integrity within LLM-driven systems. We design a self-calibration mechanism that mitigates drift and dynamically approximates ground truth by incorporating an ARIMA forecaster, without requiring continuous human oversight. To demonstrate the effectiveness and reliability of our methodology, we apply it to the complex domain of profiling the resource usage of zero-knowledge workloads in edge computing networks. Experimental results show that the proposed self-calibrating agentic framework successfully profiles the zero-knowledge workloads, achieving a higher accuracy than baseline LLM agents by 91.7% for resource usage prediction and improving the prediction speed by 71.7% compared to pure profiling, establishing a robust foundation for deploying autonomous AI in decentralized infrastructures. Furthermore, the ground truth generation using the proposed ARIMA leaping algorithm is 52% faster than a standard ARIMA forecasting algorithm, while achieving the same accuracy.

cs.NI↗

arXiv2026

Optimizing Split Learning Latency in TinyML-Based IoT Systems

Split learning (SL) addresses the limitation of running deep learning inference directly on low-power edge/IoT nodes, in which it executes part of the inference process on the sensor and offloading the remainder to a companion device. Despite its promise, the inference latency of SL on constrained hardware under realistic low-power wireless protocols remains unexplored. This paper presents the first experimental latency benchmark of TinyML-based SL on ESP32-S3 boards, comparing four wireless communication protocol solutions (UDP, TCP, ESP-NOW, BLE). We also analyze the impact of the choice of different split points across different models (MobileNet-V2 and ResNet50) in terms of communication and computation overhead as a way to minimize the end-to-end inference latency. We propose a Beam Search-based algorithm for split point optimization that minimizes end-to-end latency, and compare it with other methods, including Greedy Search, First-Fit, Random-Fit, and Brute Force. ESP-NOW achieves the best RTT (3.6 s) and serves as the base protocol for the algorithm, which delivers near-optimal latency with processing time of 0.1 s for 5 devices.

cs.NI↗

arXiv2026

Efficient Self-Learning and Model Versioning for AI-native O-RAN Edge

The AI-native vision of 6G requires Radio Access Networks to train, deploy, and continuously refine thousands of machine learning (ML) models that drive real-time radio network optimization. Although the Open RAN (O-RAN) architecture provides open interfaces and an intelligent control plane, it leaves the life-cycle management of these models unspecified. Consequently, operators still rely on ad-hoc, manual update practices that can neither scale across the heterogeneous, multi-layer stack of Cell-Site, Edge-, Regional-, and Central-Cloud domains, nor across the three O-RAN control loops (real-, near-real-, and non-real-time). We present a self-learning framework that provides an efficient closed-loop version management for an AI-native O-RAN edge. In this framework, training pipelines in the Central/Regional Cloud continuously generate new models, which are cataloged along with their resource footprints, security scores, and accuracy metrics in a shared version repository. An Update Manager consults this repository and applies a self-learning policy to decide when and where each new model version should be promoted into operation. A container orchestrator then realizes these decisions across heterogeneous worker nodes, enabling multiple services (rApps, xApps, and dApps) to obtain improved inference with minimal disruption. Simulation results show that an efficient RL-driven decision-making can guarantee quality of service, bounded latencies while balancing model accuracy, system stability, and resilience.

cs.NI↗

arXiv2026

A Cross-Layer Analysis of Network Antifragility with RIS-assisted Links under Jamming Attacks

Antifragility is an economics term defined as measure of (monetary) benefits gained from the adverse events and variability of the markets. This paper integrates for the first time the antifragility into the network based on communication links with Reconfigurable Intelligent Surface (RIS) affected by a jamming attack. We analyze whether antifragility can be achieved for several jamming models. Beyond the link-level gains, the results reveal how antifragile RIS-assisted links can be integrated into multi-hop systems to improve end-to-end network resilience, connectivity, and throughput under adversarial effects.

cs.NI↗

arXiv2025

A Secured Intent-Based Networking (sIBN) with Data-Driven Time-Aware Intrusion Detection

While Intent-Based Networking (IBN) promises operational efficiency through autonomous and abstraction-driven network management, a critical unaddressed issue lies in IBN's implicit trust in the integrity of intent ingested by the network. This inherent assumption of data reliability creates a blind spot exploitable by Man-in-the-Middle (MitM) attacks, where an adversary intercepts and alters intent before it is enacted, compelling the network to orchestrate malicious configurations. This study proposes a secured IBN (sIBN) system with data driven intrusion detection method designed to secure legitimate user intent from adversarial tampering. The proposed intent intrusion detection system uses a ML model applied for network behavioral anomaly detection to reveal temporal patterns of intent tampering. This is achieved by leveraging a set of original behavioral metrics and newly engineered time-aware features, with the model's hyperparameters fine-tuned through the randomized search cross-validation (RSCV) technique. Numerical results based on real-world data sets, show the effectiveness of sIBN, achieving the best performance across standard evaluation metrics, in both binary and multi classification tasks, while maintaining low error rates.

cs.CR↗

arXiv2025

On Efficient Topology Management in Service-Oriented 6G Networks: An Edge Video Distribution Case Study

An efficient topology management in future 6G networks is one of the fundamental challenges for a dynamic network creation based on location services, whereby each autonomous network entity, i.e., a sub-network, can be created for a specific application scenario. In this paper, we study the performance of a novel topology changes management system in a sample 6G network being dynamically organized in autonomous sub-networks. We propose and analyze an algorithm for intelligent prediction of topology changes and provide a comparative analysis with topology monitoring based approach. To this end, we present an industrially relevant case study on edge video distribution, as it is envisioned to be implemented in line with the 3GPP and ETSI MEC (Multi-access Edge Computing) standards. For changes prediction, we implement and analyze a novel topology change prediction algorithm, which can automatically optimize, train and, finally, select the best of different machine learning models available, based on the specific scenario under study. For link change scenario, the results show that three selected ML models exhibit high accuracy in detecting changes in link delay and bandwidth using measured throughput and RTT. ANN demonstrates the best performance in identifying cases with no changes, slightly outperforming random forest and XGBoost. For user mobility scenario, XGBoost is more efficient in learning patterns for topology change prediction while delivering much faster results compared to the more computationally demanding deep learning models, such as LSTM and CNN. In terms of cost efficiency, our ML-based approach represents a significantly cost-effective alternative to traditional monitoring approaches.

cs.NI↗

arXiv2024

Optimizing LoRa for Edge Computing with TinyML Pipeline for Channel Hopping

We propose to integrate long-distance LongRange (LoRa) communication solution for sending the data from IoT to the edge computing system, by taking advantage of its unlicensed nature and the potential for open source implementations that are common in edge computing. We propose a channel hoping optimization model and apply TinyML-based channel hoping model based for LoRa transmissions, as well as experimentally study a fast predictive algorithm to find free channels between edge and IoT devices. In the open source experimental setup that includes LoRa, TinyML and IoT-edge-cloud continuum, we integrate a novel application workflow and cloud-friendly protocol solutions in a case study of plant recommender application that combines concepts of microfarming and urban computing. In a LoRa-optimized edge computing setup, we engineer the application workflow, and apply collaborative filtering and various machine learning algorithms on application data collected to identify and recommend the planting schedule for a specific microfarm in an urban area. In the LoRa experiments, we measure the occurrence of packet loss, RSSI, and SNR, using a random channel hoping scheme to compare with our proposed TinyML method. The results show that it is feasible to use TinyML in microcontrollers for channel hopping, while proving the effectiveness of TinyML in learning to predict the best channel to select for LoRa transmission, and by improving the RSSI by up to 63 %, SNR by up to 44 % in comparison with a random hopping mechanism.

cs.NI↗

arXiv2024

Towards Smart Microfarming in an Urban Computing Continuum

Microfarming and urban computing have evolved as two distinct sustainability pillars of urban living today. In this paper, we combine these two concepts, while majorly extending them jointly towards novel concepts of smart microfarming and urban computing continuum. Smart microfarming is proposed with applications of artificial intelligence (AI) in microfarming, while an urban computing continuum is proposed as a major extension of the concept towards an efficient Internet of Things (IoT) -edge-cloud continuum. We propose and build a system architecture for a plant recommendation system that uses machine learning (ML) at the edge to find, from a pool of given plants, the most suitable ones for a given microfarm using monitored soil values obtained from IoT sensor devices. Moreover, we propose to integrate long-distance LongRange (LoRa) communication solution for sending the data from IoT to the edge system, due to its unlicensed nature and potential for open source implementations. Finally, we propose to integrate open source and less constrained application protocol solutions, such as Advanced Message Queuing Protocol (AMQP) and Hypertext Transport Protocol (HTTP) protocols, for storing the data in the cloud. An experimental setup is used to evaluate and analyze the performance and reliability of the data collection procedure and the quality of the recommendation solution. Furthermore, collaborative filtering is used for the completion of an incomplete information about soils and plants. Finally, various ML algorithms are applied to identify and recommend the optimal plan for a specific microfarm in an urban area.

cs.NI↗

arXiv2024

Effective ML Model Versioning in Edge Networks

Machine learning (ML) models, data and software need to be regularly updated whenever essential version updates are released and feasible for integration. This is a basic but most challenging requirement to satisfy in the edge, due to the various system constraints and the major impact that an update can have on robustness and stability. In this paper, we formulate for the first time the ML model versioning optimization problem, and propose effective solutions, including the update automation with reinforcement learning (RL) based algorithm. We study the edge network environment due to the known constraints in performance, response time, security, and reliability, which make updates especially challenging. The performance study shows that model version updates can be fully and effectively automated with reinforcement learning method. We show that for every range of server load values, the proper versioning can be found that improves security, reliability and/or ML model accuracy, while assuring a comparably lower response time.

cs.NI↗

arXiv2024

Signaling Rate and Performance of RIS Reconfiguration and Handover Management in Next Generation Mobile Networks

We consider the problem of signaling rate and performance for an efficient control and management of RIS reconfigurations and handover in next generation mobile networks. To this end, we first analytically determine the rates of RIS reconfigurations and handover using a stochastic geometry network model. We derive closed-form expressions of these rates while taking into account static obstacles (both known and unknown), self-blockage, RIS location density, and variations in the angle and direction of user mobility. Based on the rates derived, we analyze the signaling rates of a sample novel signaling protocol, which we propose as an extension of an handover signaling protocol standard in mobile networks. The results quantify the impact of known and unknown obstacles on the RIS and handover reconfiguration rate as function of device density and mobility. We use the proposed analysis to evaluate the signaling overhead due to RIS reconfigurations, as well as to dimension the related RIS control plane server capacity in the network management system. To the best of our knowledge, this is the first analytical model to derive the closed form expressions of RIS reconfiguration rates, along with handover rates, and relate its statistical properties to the signaling rate and performance in next generation mobile networks.

cs.NI↗

arXiv2024

Outage Probability Analysis of Wireless Paths with Faulty Reconfigurable Intelligent Surfaces

We consider a next generation wireless network incorporating a base station a set of typically low-cost and faulty Reconfigurable Intelligent Surfaces (RISs). The base station needs to select the path including the RIS to provide the maximum signal-to-noise ratio (SNR) to the user. We study the effect of the number of elements, distance and RIS hardware failure on the path outage probability, and based on the known signal propagation model at high frequencies, derive the closed-form expression for the said probability of outage. Numerical results show the path outage likelihood as function of the probability of hardware failure of RIS elements, the number of elements, and the distance between mobile users and the RIS.

cs.IT↗

arXiv2024

Predictive Intent Maintenance with Intent Drift Detection in Next Generation Network

Intent-Based Networking (IBN) is a known concept for enabling the autonomous configuration and self-adaptation of networks. One of the major issues in IBN is maintaining the applied intent due the effects of drifts over time, which is the gradual degradation in the fulfillment of the intents, before they fail. Despite its critical role to intent assurance and maintenance, intent drift detection was largely overlooked in the literature. To fill this gap, we propose an intent drift detection algorithm for predictive maintenance of intents which can use various unsupervised learning techniques (Affinity Propagation, DBSCAN, Gaussian Mixture Models, Hierarchical clustering, K-Means clustering, OPTICS, One-Class SVM), here applied and comparatively analyzed due to their simplicity, yet efficiency in detecting drifts. The results show that DBSCAN is the best model for detecting the intent drifts. The worst performance is exhibited by the Affinity Propagation model, reflected in its poorest accuracy and latency values.

cs.NI↗

arXiv2023

Towards Optimal Serverless Function Scaling in Edge Computing Network

Serverless computing has emerged as a new execution model which gained a lot of attention in cloud computing thanks to the latest advances in containerization technologies. Recently, serverless has been adopted at the edge, where it can help overcome heterogeneity issues, constrained nature and dynamicity of edge devices. Due to the distributed nature of edge devices, however, the scaling of serverless functions presents a major challenge. We address this challenge by studying the optimality of serverless function scaling. To this end, we propose Semi-Markov Decision Process-based (SMDP) theoretical model, which yields optimal solutions by solving the serverless function scaling problem as a decision making problem. We compare the SMDP solution with practical, monitoring-based heuristics. We show that SMDP can be effectively used in edge computing networks, and in combination with monitoring-based approaches also in real-world implementations.

cs.NI↗

arXiv2023

Scaling Serverless Functions in Edge Networks: A Reinforcement Learning Approach

With rapid advances in containerization techniques, the serverless computing model is becoming a valid candidate execution model in edge networking, similar to the widely used cloud model for applications that are stateless, single purpose and event-driven, and in particular for delay-sensitive applications. One of the cloud serverless processes, i.e., the auto-scaling mechanism, cannot be however directly applied at the edge, due to the distributed nature of edge nodes, the difficulty of optimal resource allocation, and the delay sensitivity of workloads. We propose a solution to the auto-scaling problem by applying reinforcement learning (RL) approach to solving problem of efficient scaling and resource allocation of serverless functions in edge networks. We compare RL and Deep RL algorithms with empirical, monitoring-based heuristics, considering delay-sensitive applications. The simulation results shows that RL algorithm outperforms the standard, monitoring-based algorithms in terms of total delay of function requests, while achieving an improvement in delay performance by up to 50%.

cs.NI↗

arXiv2022

Towards Optimal Path Allocation for Unreliable Reconfigurable Intelligent Surfaces

Terahertz (THz) communications and reconfigurable intelligent surfaces (RISs) have been recently proposed to enable various powerful indoor applications, such as wireless virtual reality (VR). For an efficient servicing of VR users, an efficient THz path allocation solution becomes a necessity. Assuming the RIS component is the most critical one in enabling the service, we investigate the impact of RIS hardware failure on path allocation performance. To this end, we study a THz network that employs THz operated RISs acting as base stations, serving VR users. We propose a Semi-Markov decision Process (SMDP)-based path allocation model to ensure the reliability of THz connection, while maximizing the total long-term expected system reward, considering the system gains, costs of link utilization, and the penalty of RIS failure. The SMDP-based model of the RIS system is formulated by defining the state space, action space, reward model, and transition probability distribution. We propose an optimal iterative algorithm for path allocation that decides the next action at each system state. The results show the average reward and VR service blocking probability under different scenarios and with various VR service arrivals and RIS failure rates, as first step towards feasible VR services over unreliable THz RIS.

cs.NI↗

arXiv2021

Benchmarking Machine Learning Techniques for THz Channel Estimation Problems

Terahertz communication is one of the most promising wireless communication technologies for 6G generation and beyond. For THz systems to be practically adopted, channel estimation is one of the key issues. We consider the problem of channel modeling and estimation with deterministic channel propagation and the related physical characteristics of THz bands, and benchmark various machine learning algorithms to estimate THz channel, including neural networks (NN), logistic regression (LR), and projected gradient ascent (PGA). Numerical results show that PGA algorithm yields the most promising performance at SNR=0 dB with NMSE of -12.8 dB.

cs.NI↗

arXiv2021

Benchmarking Buffer Size in IoT Devices Deploying REST HTTP

A few potential IoT communication protocols at the application layer have been proposed, including MQTT, CoAP and REST HTTP, with the latter being the protocol of choice for software developers due to its compatibility with the existing systems. We present a theoretical model of the expected buffer size on the REST HTTP client buffer in IoT devices under lossy wireless conditions, and validate the study experimentally. The results show that increasing the buffer size in IoT devices does not always improve performance in lossy environments, hence demonstrating the importance of benchmarking the buffer size in IoT systems deploying REST HTTP.

cs.NI↗

arXiv2021

Benchmarking Various ML Solutions in Complex Intent-Based Network Management Systems

Intent-based networking (IBN) solutions to managing complex ICT systems have become one of the key enablers of intelligent and autonomous network management. As the number of machine learning (ML) techniques deployed in IBN increases, it becomes increasingly important to understand their expected performance. Whereas IBN concepts are generally specific to the use case envisioned, the underlying platforms are generally heterogenous, comprised of complex processing units, including CPU/GPU, CPU/FPGA and CPU/TPU combinations, which needs to be considered when running the ML techniques chosen. We focus on a case study of IBNs in the so-called ICT supply chain systems, where multiple ICT artifacts are integrated in one system based on heterogeneous hardware platforms. Here, we are interested in the problem of benchmarking the computational performance of ML technique defined by the intents. Our benchmarking method is based on collaborative filtering techniques, relying on ML-based methods like Singular Value Decomposition and Stochastic Gradient Descent, assuming initial lack of explicit knowledge about the expected number of operations, framework, or the device processing characteristics. We show that it is possible to engineer a practical IBN system with various ML techniques with an accurate estimated performance based on data from a few benchmarks only.

cs.NI↗