Research map / Federated learning and privacy / Communication-efficient FL
Unlocking the Potential of Weighting Methods in Federated Learning Through Communication Compression
Valery Parfenov, Nail Bashirov, Daniil Medyakov, Dmitry Bylinkin, Aleksandr Beznosikov
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 |
|---|---|---|---|
| use ADI joint weighting-compression instead of DIANA | federated compression · CIFAR-10 Dirichlet 0.3, Rand 5% | max accuracy (%): 75.4 -> 82.4 | Improved |
| use ADI instead of EF21 | federated compression · CIFAR-10 Dirichlet 0.3, Rand 5% | max accuracy (%): 61.6 -> 82.4 | Improved |
| use ADI instead of DIANA on CIFAR-100 | federated compression · CIFAR-100 Dirichlet 0.5, Rand 10% | max accuracy (%): 70.1 -> 71.7 | Improved |
| use ADI instead of DIANA at Rand 5% on CIFAR-100 | federated compression · CIFAR-100 Dirichlet 0.5, Rand 5% | max accuracy (%): 71.1 -> 70.9 | No clear effect |
| use ADI instead of AFL with naive gradient compression | weighting methods · CIFAR-10 Dirichlet 0.5, Rand 5%, p=1.0 | max accuracy (%): 47.2 -> 84.9 | Improved |
| use ADI instead of q-FFL with naive gradient compression | weighting methods · CIFAR-100 Dirichlet 0.5, Rand 5%, p=1.0 | max accuracy (%): 41.5 -> 70.9 | Improved |
| lower client participation probability p from 1.0 to 0.5 | ADI · CIFAR-10 Dirichlet 0.5, Rand 5% | max accuracy (%): 84.9 -> 79.3 | Hurt |
| increase compression from Rand 50% to Rand 5% | ADI · CIFAR-10 Dirichlet 0.5, p=1.0 | max accuracy (%): 85.2 -> 84.9 | No clear effect |
| increase compression from Rand 50% to Rand 5% | AFL · CIFAR-10 Dirichlet 0.5, p=1.0 | max accuracy (%): 67.2 -> 47.2 | Hurt |
| add agnostic weighting heuristically to baselines (DIANA, EF21) (ablation) | DIANA and EF21 · CIFAR-10 non-i.i.d., Rand 10% | accuracy vs communication rounds (Figure 2) | Improved |
| add compression to ADI (versus no compression) (ablation) | ADI · CIFAR-10 non-i.i.d. | accuracy vs communication rounds (Figure 1) | No clear effect |
Closest papers by meaning
- Stochastic Controlled Averaging for Federated Learning with Communication Compression (ICLR 2024 · similarity 0.88)
- EFSkip: A New Error Feedback with Linear Speedup for Compressed Federated Learning with Arbitrary Data Heterogeneity (AAAI 2025 · similarity 0.87)
- FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025 · similarity 0.87)
- The Panaceas for Improving Low-Rank Decomposition in Communication-Efficient Federated Learning (ICML 2025 · similarity 0.87)
- SILVER: Single-loop variance reduction and application to federated learning (ICML 2024 · similarity 0.86)
- Interaction-Aware Gaussian Weighting for Clustered Federated Learning (ICML 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.
