New framework simplifies federated domain adaptation with knowledge distillation

Researchers have introduced SKD-ETR, a novel framework for one-shot federated unsupervised domain adaptation. This method addresses challenges in existing federated learning approaches by reducing communication rounds and improving robustness against noisy labels. SKD-ETR employs Smoothed Knowledge Distillation to train a student model on a target domain using smoothed pseudo-labels from source models, while Exponential moving average-based Teacher Refinement transfers knowledge back to the source models. Experiments on benchmark datasets like OfficeHome and DomainNet show SKD-ETR achieves competitive performance with enhanced efficiency and robustness. AI

IMPACT Simplifies federated learning for domain adaptation, potentially improving efficiency and robustness in decentralized AI systems.

RANK_REASON The cluster contains a research paper detailing a new method for domain adaptation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• DomainNet
• Exponential moving average-based Teacher Refinement
• Federated unsupervised domain adaptation
• Office-Caltech
• OfficeHome
• SKD-ETR
• Smoothed Knowledge Distillation

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →