Editing Federated Learning (section)

== 5.3.2 Communication Efficiency in Edge-Based FL ==

Communication remains one of the most critical bottlenecks in deploying FL at the edge, where devices often suffer from limited bandwidth, intermittent connectivity, and energy constraints. To address this, several strategies have been developed:

* '''Gradient quantization''': Reduces the size of transmitted updates by lowering numerical precision (e.g., from 32-bit to 8-bit values).
* '''Gradient sparsification''': Limits communication to only the most significant changes in the model, transmitting top-k updates while discarding negligible ones.
* '''Local update batching''': Allows devices to perform multiple rounds of local training before sending updates, reducing the frequency of synchronization.

Further, client selection strategies dynamically choose a subset of devices to participate in each round, based on criteria like availability, data quality, hardware capacity, or trust level. These communication optimizations are crucial for ensuring that FL remains scalable, efficient, and deployable across millions of edge nodes without overloading the network or draining device batteries.<sup>[1]</sup><sup>[2]</sup><sup>[3]</sup>