Edge 279: Cross-Silo Federating Learning

Cross-silo federated learning(FL), Amazon’s research on personalized FL and IBM’s FL framework.

Created Using Midjourney

In this issue:

1. Explore cross-silo federated learning(FL)

2. Amazon’s research on personalized FL and

3. IBM’s federated learning framework.

💡 ML Concept of the Day: Cross-Silo Federated Learning

In the last few editions of our series about federated learning, we have discussed architectures based on data partitioning structure such as horizontal, vertical, and transfer federated learning. An alternative group of architectures are based on the architecture scheme. The two most relevant architectures in this group are known as cross-silo and cross-device federated learning.

Cross-silo federated learning(CSFL) , represents an extension of the original federated learning architecture in which the nodes are segregated into silos. Instead of all nodes sharing the model updates with a central server, they can join the federation through a specific silo that segregates storage and compute and a localized version of the model. The updates from the different silos are then shared to improve the global model.