Deep Learning-Based Dynamic Computation Task Offloading for Mobile Edge Computing Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Mobile Edge Computing (MEC) networks face challenges with dynamic weighted tasks.
• Existing deep learning methods for computation offloading require extensive retraining for environmental changes.

Purpose of the Study:

• To develop an adaptive computational offloading algorithm for MEC networks.
• To minimize system utility by optimizing offloading decisions and bandwidth allocation.

Main Methods:

• Proposed a deep supervised learning-based computational offloading (DSLO) algorithm.
• Incorporated batch normalization to enhance model convergence and robustness.
• Formulated the joint optimization as a mixed-integer programming (MIP) problem.

Main Results:

• DSLO demonstrates rapid adaptation to new MEC scenarios with minimal training samples.
• Achieved 99% normalized system utility using only four training samples per scenario.
• Outperforms traditional methods in adaptability and efficiency.

Conclusions:

• DSLO offers a robust and efficient solution for computation offloading in dynamic MEC environments.
• Enables faster deployment of offloading algorithms in future MEC networks.
• Highlights the potential of deep supervised learning for adaptive network management.