Federated Learning: How Intelligent Agents Contribute to Collaborative AI
Problem Statement:
Learning in a distributed way can be really hard. Federated Learning is a branch of Machine Learning in which different devices learn locally and share their models to achieve a global objective. This research aims to explore the role of intelligent agents in federated learning, focusing on the challenges and solutions related to sharing information and Non-IID (Non-Independent and Identically Distributed) learning in networked systems. Federated learning enables multiple agents to collaboratively train machine learning models while preserving data privacy. However, the heterogeneity of data across agents and the need for efficient information sharing pose significant challenges. The study will investigate how intelligent agents can enhance federated learning by addressing these issues and improving the overall performance and robustness of collaborative AI systems. We want to look at Personalized Federated Learning (PFL): PFL aims to create personalized models for each client by combining global and local model updates. Techniques like Meta-Learning and Transfer Learning can be used to enhance personalization.
Tasks:
We have different dataset for non-IID data and the objective is to build a PFL solution using the Flower framework. So the project entails a literature review, a problem selection (preferably in the IoT and Sensing fields) and the implementation in the framework.
Work:
10% Theory, 70% Simulations, 20%Writing
Contact:
Alessandro Chiumento (a.chiumento@utwente.nl)