Metropolis-Hastings with Levy jumps prevents entrapment in weighted random walks, yielding a convergence rate that accounts for data heterogeneity, network spectral gap, and jump probability.
Improved generalization bounds for communication efficient federated learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Decentralized Learning via Random Walk with Jumps
Metropolis-Hastings with Levy jumps prevents entrapment in weighted random walks, yielding a convergence rate that accounts for data heterogeneity, network spectral gap, and jump probability.