TrajSyn: Privacy-Preserving Dataset Distillation for Federated Model Training

The paper presents TrajSyn, a novel method for distilling datasets in a privacy-preserving manner, crucial for server-side adversarial training within federated learning environments. The research addresses a critical challenge in secure and robust AI, particularly in scenarios where data privacy is paramount.

• •Addresses privacy concerns in federated learning by distilling data trajectories.
• •Focuses on server-side adversarial training, improving model robustness.
• •Offers a method to balance model performance and data privacy.