Protection of Neural Network Models from Privacy Breach Threats in Federated Learning Using Optimization Methods

Key Points

• The central aim is to address privacy breach threats in federated learning using optimization techniques.
• Utilized optimization methods to transform weights of local neural network models.
• Developed a mechanism for creating new weights for secure transmission.
• Conducted experimental studies to evaluate the approach's effectiveness.
• Confirmed the effectiveness of the developed optimization method in preventing privacy breaches.
• Demonstrated that transformed weights reduce the risk of interception by attackers.