Multi-arm bandit-led clustering in federated learning

• Main Author: Zhao, Joe Chen Xuan
• Other Authors: Anupam Chattopadhyay
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Multi-arm bandits
• Online Access: https://hdl.handle.net/10356/175424
• Institution: Nanyang Technological University

Federated Learning (FL) is a machine learning technique that enables the training of models across decentralized devices or nodes, without requiring the raw data to be centrally collected in one location. Instead, the model is trained in a distributed manner across multiple nodes, with each node only sending the model updates (and not the raw data) to a central server. The project’s direction was to explore and train an agent capable of recognizing which node can contribute best to maximize an existing cluster’s federated learning accuracy. The factor that was studied in this project was noise introduced to the data of a certain node that alters the data quality. The outcomes of the project showed that using reinforcement learning to train an agent that is capable of selecting the best node significantly improves the federated accuracy. As well as some noise alterations do make the model more robust in some cases.