OrchNAS: Orchestrated Neural Architecture Search Service for Personalised Federated Edge Intelligence

OrchNAS presents an energy-aware, personalised federated edge intelligence framework for heterogeneous devices. Its server-side Neural Architecture Search Service learns a compact global architecture representation across services, then derives service-specific subnets through progressive greedy pruning under device energy, computation, and memory constraints. A primal-dual optimisation scheme enforces energy budgets while adapting personalised parameters without discarding the global representation. The abstract reports experiments on real-world and benchmark datasets, but does not provide dataset names, quantitative results, baselines, or ablations.