FedPromo: Federated Lightweight Proxy Models at the Edge Bring New Domains to Foundation Models
The episode discusses 'FedPromo,' a framework that combines knowledge distillation and federated learning to make high-end AI accessible at the edge. The authors explain a two-stage process involving server-side alignment of proxy models to foundation models, followed by client deployment of lightweight classifiers. Key innovations like ICP and CDB address challenges from non-IID data distribution, resulting in robust, privacy-preserving decentralized intelligence.