IntraShuffler: A Privacy Preserving Framework for Heterogeneous DP Federated Learning
- Published
- Source
- arXiv
- Paper number
- 296
- Field
- Machine Learning
- arXiv ID
- 2606.02563
Key points
- Heterogeneous differential privacy (HDP) in federated learning allows clients to choose individual privacy budgets and supports budget-aware aggregation.
- Experiments across four different datasets show that IntraShuffler reduces gradient reconstruction risk by more than 60% and lowers surrogate inference accuracy from 0.78 to 0.33, while maintaining similar model utility across multiple FL aggregation rules.
Paper links
External research summaries. These are not HDATF publications or measured product results.