Applications of ML in Networking and IoT - LAAS - Laboratoire d'Analyse et d'Architecture des Systèmes Access content directly
Master Thesis Year : 2020

Applications of ML in Networking and IoT


The emergence of Machine Learning (ML) has increased exponentially in numerous appli-cations, including computer vision, speech recognition and natural language processing. Thisis due to the intersection of the massive data generated by IoT devices (e.g. sensors, mobilephones and smart devices) and the computing power available in cloud and data centers. Hence,complex systems have been designed to improve the quality of life such as health monitoring,smart homes and intelligent transport systems. However, the cloud computing has a number ofissues related to the trade-off between efficiency and latency. Indeed, the former cannot ensurethe performance for real-time applications due to the long-range network latency between thecloud and end devices. For that reason, we propose to leverage these issues in this work byproposing alternative solutions. First, we present a real-time network traffic prediction usingdifferent DL models and compare their performance to a classic ML model named SVR byusing the Hyperparameters tuning process. In fact, traffic prediction is a tool for Internetservice providers to serve the customers better QoS, i.e. lessen the latency for sensitive applica-tions. Second, we study the edge computing paradigm that attempts to unleash DL services bybringing the cloud capabilities close to end users by deploying edge servers. Thus, we propose adistributed inference framework for DL models. The latter intends to attain a high performancewhile minimizing the computation latency for sensitive services and the energy consumptionof IoT devices. Extensive evaluations on the standard CNN "ResNet20" for object recognitiontask using CIFAR10 dataset shows the effectiveness of the proposed framework in controllinglatency versus inference accuracy. To further enhance the distributed inference framework, wepropose automating the offloading decision and resource allocation of computing tasks for IoTdevices using Deep Q-Network (DQN), a well-known DRL algorithm. Several Simulations showthat DQN has been successful in managing execution of tasks belongs to IoT devices in orderto make the best use of available resources.
Fichier principal
Vignette du fichier
Rapport PFE.pdf (2.27 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-02932494 , version 1 (07-09-2020)


  • HAL Id : hal-02932494 , version 1


Saif Eddine Nouma. Applications of ML in Networking and IoT. Artificial Intelligence [cs.AI]. 2020. ⟨hal-02932494⟩
39 View
77 Download


Gmail Facebook X LinkedIn More