Abstract: Continuously orchestrating in-network model aggregations for federated learning faces fundamental challenges such as the combinatorial nature of traffic reduction, the dynamic trade-offs ...