M-MIMO Channel Estimation using Federated Learning

examine the feasibility of an innovative solution that combines Federated Learning with Deep Reinforcement Learning (DRL) to optimize the selection of edge nodes for model aggregation. Through the proposed DRL-based client selection method, the system can actively select nodes that have shown superior performance based on their training quality, computational resources, and network conditions. This selection mechanism continuously interacts with the environment, learning and adapting to changes in network conditions and client states. The presented architecture employs DRL to screen edge node behavior and to facilitate the selection of nodes by considering multiple factors such as computational power, latency, data quality, and more. The client selection algorithm utilizes DRL to transform the device node selection problem into a Markov Decision Process (MDP), defining comprehensive action and state spaces as well as a reward function that is aligned with the primary objectives of efficient and effective FL. The customized reward function enables the RL agent to achieve faster model convergence. This innovation holds immense potential for diverse applications and is poised to be a game-changer in the IoT domain.

Faculty Supervisor:

Wei Shi

Student:

Partner:

Ericsson Canada Inc (Ottawa, ON)

Discipline:

Computer science

Sector:

Information and cultural industries; Manufacturing; Professional, scientific and technical services

University:

Carleton University

Program: