Research map / Federated learning and privacy / Communication-efficient FL
Federated Automatic Differentiation
Keith Rush, Zachary Charles, Zachary Garrett
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn server learning rate and momentum by federated hypergradient descent (ablation) | FedAvgM · SO NWP (default server init, random client LR) | max test accuracy (%): 25.0 -> 25.7 | Improved |
| learn server learning rate and momentum by federated hypergradient descent (ablation) | FedAvgM · Shakespeare (default server init, random client LR) | max test accuracy (%): 56.8 -> 57.3 | Improved |
| learn server learning rate and momentum by federated hypergradient descent (ablation) | FedAvgM · EMNIST (default server init, random client LR) | max test accuracy (%): 87.0 -> 87.1 | Improved |
| learn server learning rate and momentum by federated hypergradient descent (ablation) | FedAvgM · CIFAR-100 (default server init, random client LR) | max test accuracy (%): 43.2 -> 45.6 | Improved |
| learn server learning rate and momentum by federated hypergradient descent (ablation) | FedAvgM · SO TP (recall@5) (default server init, random client LR) | max test accuracy (%): 32.2 -> 35.8 | Improved |
| learn server hyperparameters instead of fixing them at defaults (default server/default client) (ablation) | FedAvgM on SO NWP · SO NWP | runs diverged early: 23/50 -> none | Improved |
| use learned hyperparameters to reach unlearned max accuracy with fewer trials | FedAvgM · bootstrap over 50 trials | trials needed: one-tenth the trials | Improved |
| learn client weighting exponent instead of example weighting (ablation) | FedAvg · CIFAR-100, EMNIST, Shakespeare, SO NWP, SO TP | test performance | No clear effect |
| learn client weighting exponent instead of fixed example or uniform weighting (ablation) | FedAvg · Synthetic(1,1) | loss | Improved |
| initialize learned weighting exponent at q=0 instead of q=1 (ablation) | FedAvg with learned weighting · Synthetic(1,1) | loss | Hurt |
Closest papers by meaning
- Privacy-Preserving Federated Convex Optimization: Balancing Partial-Participation and Efficiency via Noise Cancellation (ICML 2025 · similarity 0.88)
- Private and Federated Stochastic Convex Optimization: Efficient Strategies for Centralized Systems (ICML 2024 · similarity 0.87)
- FedLMT: Tackling System Heterogeneity of Federated Learning via Low-Rank Model Training with Theoretical Guarantees (ICML 2024 · similarity 0.87)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 2025 · similarity 0.86)
- Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning (AAAI 2025 · similarity 0.86)
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
