How much FL and edge AI must be demonstrated at proof-of-concept stage?
How much of the federated learning and edge AI side does the program want to actually see working at the proof-of-concept stage? We could stand both up at small scale early, or start with a centralized cloud baseline and layer them in later. Curious which the program prefers.
Atharv Arya
shared this idea
-
A simulated federation is sufficient for a Phase I architecture demonstration; a live multi-party federated learning deployment involving distinct legal entities is not required until Phase II. For Phase I, the primary deliverable is to demonstrate a functional predictive AI architecture and build a preliminary model (proof of concept) for a selected freight corridor or segment using combined real-world and synthetic datasets. The actual implementation of federated capabilities to enable distributed model learning and real-time processing (while preserving data privacy) is identified as a Phase II outcome.
