Two-step personalized federated learning algorithm in reality

Machine learning attempts to model high-level abstractions in data using multiple processing layers with complex structures or non-linear transformations. Federated learning is a distributed machine learning approach that allows multiple parties to collaborate on training while preserving user data privacy. However, the data from each party is typically non-independent and identically distributed (Non-IID), which can negatively impact the training effectiveness of the model. This study proposes a contrastive learning method to mitigate the impact of Non-IID data distribution on model training. Additionally, this study researches the feasibility of deploying this method on edge devices, for example, the Internet of Things (IoT). The primary objective of this research project is to demonstrate combining contrastive learning with clustering methods. It can solve the impact caused by Non-IID distribution in federated learning and produce models that balance both generality and personalization. The study aims to validate the research methods through more diverse datasets and data distributions closer to reality.

Faculty Supervisor:

Patrick Hung

Student:

Partner:

National Cheng Kung University

Discipline:

Computer science

Sector:

Artificial Intelligence; Technology

University:

University of Ontario Institute of Technology

Program: