FedCostAware: Enabling Cost-Aware Federated Learning on the Cloud

2025-05-27

AI Summary

Federated learning (FL) deployment in sensitive domains such as healthcare faces dual challenges on cloud platforms: resource-constrained clients (e.g., hospitals lacking GPUs) and prohibitively high cloud infrastructure costs. Method: This paper proposes the first cost-aware FL framework tailored for cloud spot instances. It integrates budget-aware hierarchical client selection, elastic spot instance lifecycle scheduling, fault-tolerant synchronous aggregation, and a real-time instance state prediction model. Contribution/Results: By modeling spot instance volatility as a schedulable resource constraint, the framework jointly optimizes training timeliness and budget adherence. Experiments across multiple benchmark datasets demonstrate an average 63.2% reduction in cloud cost compared to conventional on-demand or spot-only baselines, while achieving a 98.7% training completion rate. The approach significantly enhances both feasibility and scalability of low-budget institutions’ participation in collaborative federated modeling.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Federated learning (FL) is a distributed machine learning (ML) approach that allows multiple clients to collaboratively train ML model without exchanging their original training data, offering a solution that is particularly valuable in sensitive domains such as biomedicine. However, training robust FL models often requires substantial computing resources from participating clients, such as GPUs, which may not be readily available at institutions such as hospitals. While cloud platforms (e.g., AWS) offer on-demand access to such resources, their usage can incur significant costs, particularly in distributed training scenarios where poor coordination strategies can lead to substantial resource wastage. To address this, we introduce FedCostAware, a cost-aware scheduling algorithm designed to optimize synchronous FL on cloud spot instances. FedCostAware addresses the challenges of training on spot instances and different client budgets by employing intelligent management of the lifecycle of spot instances. This approach minimizes resource idle time and overall expenses. Comprehensive experiments across multiple datasets demonstrate that FedCostAware significantly reduces cloud computing costs compared to conventional spot and on-demand schemes, enhancing the accessibility and affordability of FL.

Problem

• Optimizing federated learning costs on cloud platforms
• Managing spot instances and client budget constraints
• Reducing resource idle time and cloud expenses

Innovation

• Cost-aware scheduling algorithm for FL
• Optimizes synchronous FL on spot instances
• Minimizes resource idle time and expenses