We study collaborative machine learning (ML) across wireless devices, each
with its own local dataset. Offloading these datasets to a cloud or an edge
server to implement powerful ML solutions is often not feasible due to latency,
bandwidth and privacy constraints. Instead, we consider