OrchNAS Optimizes AI Models for Personalized Federated Edge Devices.

Deploying artificial intelligence on heterogeneous edge devices presents significant challenges due to varying energy, computational, and memory constraints. Traditional approaches often struggle to provide personalized models that are optimized for each device's unique capabilities. A new framework, OrchNAS, addresses this by introducing an orchestrated Neural Architecture Search (NAS) service for personalized federated edge intelligence. This system allows a central server to manage the architecture search process, enabling individual edge services to derive custom-tailored AI models. OrchNAS features an energy-aware global architecture search that learns a compact representation across diverse services. It then employs an energy-efficient selection mechanism, using a progressive, greedy pruning strategy, to extract a personalized sub-network for each edge device that adheres to its resource limitations. Furthermore, an energy-efficient personalized model optimization scheme updates device-specific parameters while preserving global representations, ensuring strict energy budgets are met during adaptation. Experiments on real-world and benchmark datasets demonstrate the effectiveness of this approach in optimizing AI models for constrained edge environments.