Adaptive Federated Hamiltonian Monte Carlo Proposed to Cut Communication Costs in Federated Learning
Researchers propose AFHMC, an adaptive federated Hamiltonian Monte Carlo method for Bayesian federated learning, claiming to break the sqrt(d) communication barrier in federated sampling.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-10 05:00 · arXiv stat.ML
Breaking the $\sqrt{d}$ Communication Barrier in Federated Sampling with Adaptive Hamiltonian Monte Carlo