CWL-FLOps: A Novel Method for Federated Learning Operations at Scale

Federated Learning (FL) has attracted much attention in recent years because it enables users with private data sets to train a global model collaboratively without raw data exchange. However, due to a lack of automation, researchers often struggled to develop, deploy, track, and manage all the data, steps, and configuration setup for all FL participating nodes. Federated Learning Operations (FLOps) is recently emerging in the FL community, a new methodology for developing FL systems efficiently and continuously. Some research works discussed approaches for FLOps, but only a few solutions address managing FL application scenarios from the workflow perspective. This poster proposes CWL-FLOps, a novel CWL-based method for FLOps, which can improve the flexibility of FL abstraction and fully automate the FL deployment and execution by mapping high-level descriptions onto distributed resource nodes. Our experiments demonstrate the feasibility of describing centralized and decentralized FL scenarios using CWL abstracted definitions without relying on heavily customized or external software for execution.