OrchNAS: Energy-Aware Federated Edge AI Framework
OrchNAS is a newly introduced framework that aims to create an energy-conscious, customized federated edge intelligence system. It leverages a Neural Architecture Search (NAS) Service to automatically generate models suited for various edge environments. This framework manages architecture searches through a server-side NAS service, enabling edge services to create tailored architectures based on specific energy, computation, and memory limitations of devices. It incorporates a global architecture search method that efficiently learns a compact representation across diverse services. Furthermore, an architecture selection process focused on energy efficiency allows each service to develop a personalized subnet that meets its resource requirements through a progressive, greedy pruning strategy. Lastly, an energy-efficient optimization scheme updates adaptive parameters while maintaining global representations.
Key facts
- OrchNAS is an energy-aware, personalised, federated edge intelligence framework.
- It leverages a Neural Architecture Search Service to design service-adaptive models.
- The framework orchestrates architecture search on a server-side NAS service.
- It enables edge services to derive personalised architectures under device-level constraints.
- An energy-aware global architecture search mechanism learns a compact global representation.
- An energy-efficient architecture selection mechanism uses progressive, greedy, energy-aware pruning.
- An energy-efficient personalised model optimisation scheme updates service-adaptive parameters.
- The framework preserves global representations while optimising for individual services.
Entities
—