ENERGY-EFFICIENT USER-EDGE ASSOCIATION AND RESOURCE ALLOCATION IN IOT-BASED HIERARCHICAL FEDERATED LEARNING
AuthorSAADAT, HASSAN ABDULRAHMAN
MetadataShow full item record
The proliferation of data as part of the Internet of Things (IoT) systems needs to be efficiently utilized while respecting data privacy and scalability. Edge computing is an emerging paradigm that mandates efficient processing of local data, close to where data is being collected. Such paradigm has motivated enormous research that merges computation and communication resources to explore many trade-offs that address heterogeneity of the IoT devices, while taking care of both scalability and data privacy. Federated learning (FL) is a distributed learning paradigm combining edge computing with artificial intelligence techniques. FL, compared to centralized learning (CL), preserves the data privacy of, and reduces the communication energy consumption by IoT devices, by requiring them to share locally trained machine learning models with the cloud rather than their private raw data. Hierarchical federated learning (HFL) improves FL by deploying a layer of edges that are responsible for multiple intermediate model aggregation rounds before the global aggregation is performed on the cloud. The HFL configuration alongside efficient user-edge association and resource allocation ensure more energy and communication efficient, and skewed-data robust learning scheme compared to FL. In this thesis, we assess the learning performance of the HFL framework while respecting IoT devices' limitations, such as energy budget, computational power, and storage space. First, HFL is evaluated in terms of learning performance and non-identically and independently distributed (non-iid) data handling by implementing an intrusion detection system (IDS) using the NSL-KDD dataset. Then, we formulate and solve a communication energy minimization problem that performs optimal client-edge association and resource allocation. We also implement an alternative less complex solution leveraging reinforcement learning (RL) that provides a fast user-edge association and resource allocation response in highly dynamic HFL networks. The proposed solutions are compared with several state-of-the-art client-edge association techniques, leveraging MNIST dataset. Moreover, we study the trade-off between minimizing the per-round energy consumption and Kullback-Leibler divergence (KLD) of the data distribution, and its effect on the total energy consumption.
- Computer Science & Engineering [79 items ]
Showing items related by title, author, creator and subject.
Saad H.; Mohamed A.; Elbatt T. ( IEEE , 2013 , Conference Paper)In this paper, we address the problem of distributed interference management of cognitive femtocells that share the same frequency range with macrocells using distributed multiagent Q-learning. We formulate and solve three ...
A cooperative Q-learning approach for distributed resource allocation in multi-user femtocell networks Saad H.; Mohamed A.; El Batt T. ( Institute of Electrical and Electronics Engineers Inc. , 2016 , Conference Paper)This paper studies distributed interference management for femtocells that share the same frequency band with macrocells. We propose a multi-agent learning technique based on distributed Q-learning, called subcarrier-based ...
Deep Learning for RF-Based Drone Detection and Identification: A Multi-Channel 1-D Convolutional Neural Networks Approach Allahham M.S.; Khattab T.; Mohamed A. ( Institute of Electrical and Electronics Engineers Inc. , 2020 , Conference Paper)Commercial unmanned aerial vehicles, or drones, are getting increasingly popular in the last few years. The fact that these drones are highly accessible to public may bring a range of security and technical issues to ...