Editing Federated Learning (section)

== 5.4.2 Secure Aggregation ==

Secure Aggregation, or SecAgg, is a protocol that encrypts local updates before they are sent to the server, ensuring that only the aggregated result is revealed. A typical SecAgg workflow includes:
# Each client randomly splits its model updates into multiple shares.
# These shares are exchanged among clients and the server in a way that no single party sees the entirety of any update.
# The server only obtains the sum of all client updates, rather than individual parameters.

This approach can thwart internal adversaries who might try to reconstruct local data from raw updates.<sup>[2]</sup> SecAgg is crucial for preserving confidentiality, especially in IoT-based FL systems where data privacy regulations (such as GDPR and HIPAA) prohibit raw data exposure.