Researchers have introduced FedDoG, a novel training-free framework designed to enhance one-shot federated learning (OSFL) by utilizing pre-trained diffusion models on the server side. This approach addresses the challenges of learning from heterogeneous client data with limited communication rounds by synthesizing data that aligns with individual client distributions. FedDoG significantly reduces client-side computational and communication costs by extracting compact domain representations and using a pre-trained image-conditioning adapter for visual guidance. Experiments show that FedDoG generates higher-quality synthetic images, improves classification accuracy, and drastically cuts down client upload and computation time. AI
IMPACT Enhances federated learning efficiency and data synthesis quality, potentially enabling more robust on-device AI.
RANK_REASON The cluster contains a research paper detailing a new method for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Diffusion Models
- FedDoG
- federated learning
- Hugging Face
- Image-Conditioning Adapter
- image processing method
- One-Shot Federated Learning
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →