FedACT: Scheduling Heterogeneous Devices for Concurrent Federated Learning
A research paper introduces FedACT, a novel device scheduling approach for concurrent federated learning (FL) systems. While standard FL focuses on optimizing a single task across decentralized devices, real-world applications increasingly require multiple machine learning tasks to train simultaneously on shared devices. Naively applying single-FL optimization to multi-FL systems leads to suboptimal performance due to device heterogeneity and resource inefficiency. FedACT addresses this by dynamically assigning devices to FL jobs based on an alignment scoring mechanism, aiming to minimize average job completion time (JCT). The paper is published on arXiv with ID 2605.00011.
Key facts
- FedACT is a resource heterogeneity-aware device scheduling approach.
- It is designed for multiple concurrent FL jobs on shared devices.
- The goal is to minimize average job completion time (JCT).
- It uses an alignment scoring mechanism for device assignment.
- The paper is on arXiv with ID 2605.00011.
- Standard single-FL techniques perform suboptimally in multi-FL settings.
- Device heterogeneity and resource inefficiency are key challenges.
- Real-world applications increasingly require concurrent FL tasks.
Entities
Institutions
- arXiv