Lesson 14: Datacenter-based Distributed Management

Question 1

How does Centralized Learning work?

Answer 1

• Aggregate all data to a central location
• Train
• Distribute model to all locations

Downsides:
- has a lot of data transfer costs (send the data to the centralized location to build the models, then send the models to distributed locations for use) – doing this is 53x slower
- data sovereignty

Question 2

How does Federated Learning work?

Answer 2

• Evaluate data locally
• Periodically aggregate model/parameters centrally
• Disseminate updates