Federated Learning with Imbalanced Client Participation

Speaker:  Shiqiang Wang – Exeter, United Kingdom
Topic(s):  Artificial Intelligence, Machine Learning, Computer Vision, Natural language processing

Abstract

Data is essential for machine learning, but in real-world scenarios, it is often distributed unevenly across sources with varying sizes, quality, and availability. This imbalance poses challenges to model training, potentially leading to biased outcomes. This talk addresses these challenges through the lens of federated learning (FL), a prominent use case where each client acts as an individual data source contributing to a global model. In FL, clients may participate in training at different frequencies that are unknown a priori, meaning that data from less active clients has less impact on the model, which can create further imbalance and bias. Building on our recent work, I will introduce a lightweight algorithm, FedAU, that dynamically adapts aggregation weights in the federated averaging (FedAvg) process to handle unknown and varying client participation rates effectively. By estimating optimal weights based on the participation history of clients, FedAU ensures fair representation of all data sources, mitigating biases and enhancing convergence. I will present the key ideas of our theoretical convergence analysis that connects estimation error with convergence guarantees, as well as experimental results validating the effectiveness of FedAU under various participation patterns. Afterwards, I will discuss how the principles and methodologies developed in this work can be extended beyond FL to address the general problem of learning from imbalanced data sources across various machine learning paradigms, providing new insights and tools for enhancing model performance and fairness in diverse settings.

About this Lecture

Number of Slides:  30
Duration:  60 minutes
Languages Available:  English
Last Updated:  12/05/2026

Request this Lecture

To request this particular lecture, please complete this online form.

Request a Tour

To request a tour with this speaker, please complete this online form.

All requests will be sent to ACM headquarters for review.