Federated & Private ML

DP-FedMeta: Differentially Private Federated Meta-Learning

DP-FedMeta: Differential Privacy in Federated Meta-Learning through Adaptive Clipping

Personalised federated learning with privacy guarantees that cost less accuracy, by letting the clipping threshold adapt during training.

The problem

Differential privacy protects individuals by adding noise, but fixed clipping in federated meta-learning wastes utility.

The approach

Meta-Clip, an adaptive clipping technique inside DP-FedMeta that adjusts the clipping threshold as training progresses, with convergence and privacy analysis.

The outcome

Substantial improvements in both privacy preservation and classification performance compared with fixed clipping.

In detail

Designed a privacy-aware federated meta-learning framework that addresses the utility loss introduced by differential privacy. The proposed Meta-Clip mechanism adapts clipping thresholds during training rather than relying on a fixed clipping norm, helping preserve learning capacity while maintaining formal privacy protection. The work focuses on personalised federated learning under heterogeneous and data-limited client settings.

Built with

• Federated Learning
• Differential Privacy
• Meta-Learning

More in Federated & Private ML

• IRT-FedLoRA: Reliability-Aware Federated Fine-Tuning
• ClusterGuardFL: Secure Cluster-Weighted Client Aggregation
• AR-FL: Anomaly-Resilient Federated Learning

Next project: Utility-Enhanced Differentially Private Federated Learning

By Kanishka Ranaweera, PhD, Machine Learning Engineer in Melbourne, Victoria, Australia. hello@kanishkabuilt.me