Engineering · 8 min read · April 22, 2026
Routing Optimization for Satellite Federated Learning: Tractable Boundaries
Researchers map which routing problems in orbital federated learning can be solved efficiently and which are computationally hard.
Study determines which satellite-based federated learning routing scenarios admit polynomial-time solutions versus NP-hard barriers.
- — In-orbit FL requires multi-hop relay routing between satellite clients and ground/orbital servers.
- — Model distribution phase: unicast/multicast and splittable/unsplittable flows have different tractability profiles.
- — Model collection phase: client selection and flow constraints alter computational hardness.
- — Rigorous proofs establish clear boundaries between solvable and intractable problem variants.
- — Tractable cases yield practical algorithms; intractable cases reveal fundamental complexity sources.
- — Analysis covers single and multiple model scenarios under varying network constraints.
- — Findings apply beyond satellites to any relay-based distributed learning topology.
Frequently asked
- In-orbit FL trains machine learning models across satellites using multi-hop inter-satellite links to relay data to a central server. Routing optimization determines how efficiently model updates traverse the network. Poor routing wastes bandwidth and increases latency; optimal routing minimizes communication cost and training time in bandwidth-constrained orbital environments.