Loading...

Adaptive Methods in Edge Computing

Karami, Farzan | 2024

0 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 57712 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Hossein Khalaj, Babak
  7. Abstract:
  8. Edge computing, by deploying resources closer to users, offers low-latency services, improved privacy, and reduced network load. Given the resource constraints at the edge, efficient resource management and maintaining service performance are of critical importance. Adaptive methods, by allocating resources and adjusting applications according to the current conditions, help balance performance and efficiency. One key application of edge computing is federated learning, which trains machine learning models in a distributed manner without exchanging raw data. However, factors such as limited bandwidth, low energy, and dynamic network conditions affect its performance and intensify the need for resource management. Autonomous systems like robots and self-driving cars also require significant computational resources for environmental perception and motion planning. Adaptive methods in these systems improve overall performance and reduce energy consumption. This dissertation presents adaptive methods for simultaneous optimization of resources and performance in federated learning and autonomous systems. For federated learning, a system model tailored to communication channel conditions and a model-based reinforcement learning method are proposed. These approaches aim to optimize bandwidth, power, processing frequency, and the number of local training rounds in federated learning. Simulation results demonstrate that the proposed method reduces training time with faster convergence and better interpretability. For autonomous systems, a modular optimization method has been proposed to optimize the resources of perception and motion planning components. This method not only enhances motion planning performance and reduces energy consumption but also achieves near-optimal performance as shown by simulation results
  9. Keywords:
  10. Edge Computing ; Adaptive Method ; Resource Optimization ; Federated Learning ; Autonomous System ; Model-Based Reinforcement Learning

 Digital Object List

 Bookmark

...see more