Understanding FL

This section focuses on providing a high-level technical understanding of how FL actually slots in as a solution to the problem setting described in the previous section. The goal of this section is for you to understand how FL fits as a solution, and to provide a conceptual basis that will be filled in by the subsequent chapters.

Defining FL

Federated learning is a method to synthesize global models from local models trained on the edge. FL was first developed by Google in 2016 for their Gboard application, which incorporates the context of an Android user’s typing history to suggest corrections and propose candidates for subsequent words. Indeed, this is the exact word recommendation problem discussed in the Edge inference and Edge training sections. The solution that Google produced was a decentralized training approach where an iterative process would compute model training updates at the edge, aggregating these updates to produce the global update...