Exploring investment strategies for federated learning infrastructure in medical care

Recently, federated learning has gained substantial attention in medical care where privacy-preserving cooperation among hospitals is required. However, in a real-world situation, the deployment of a federated learning system among hospitals requires heavy investment in computing and network infrastructure. Under such a case, making investment effective across computing power and network capability is essential. In this paper, we propose an investment methodology following the growth saturation of learning efficiency. We also systematically study the impacts of non-investment factors on the application of this methodology. With consideration of relevant cost models, the methodology is validated cost-effective.