Get Started
Home
Topics
Search
Library
Research questionHow can personalized federated learning reduce bidirectional model communication under bandwidth constraints without sacrificing accuracy?Personalized federated learning exchanges high-dimensional parameters among distributed devices despite heterogeneous client data and limited bandwidth. Uniform one-bit compression can lose layer-specific and fine-grained parameter information, making communication savings difficult to reconcile with accuracy.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.PACE: Propagation-Aware Collaborative Correction for One-Shot Personalized Federated Graph LearningThe study evaluates a compact propagation-aware correction that augments a complete local predictor while leaving model parameters fixed and selecting its predictive contribution from receiver validation nodes. Across six datasets, rank-6 returns use 9.6–17.6% of dense tensor bytes; the correction improves accuracy and weighted F1 on five datasets, while validation calibration assigns it zero weight on ogbn-arxiv and preserves local predictions.research paper · Sep 4, 2026Communication-Efficient Personalized Federated Learning via Layer-Wise Multi-Threshold Random SketchingThe source concerns collaborative training of client-specific models without sharing raw data, using compact low-bit sketches for bidirectional communication. It presents layer-wise multi-threshold random sketching and compares its communication-accuracy tradeoff with one-bit compression approaches; deployment hardware and additional access requirements are not specified.research paper · Sep 4, 2026
Related questions
How can federated learning personalize regional query understanding while preserving privacy and avoiding transformer training failures?How can federated learning stay reliable when heterogeneous edge clients are intermittently available and availability changes over time?How can differentially private federated learning preserve faithful explanations without sacrificing predictive utility?How can language models personalize for many users without costly per-user model storage?