Showing 2026Show all
2 papers · 1 filter
cs.LG2026
Clipping Makes Distributed and Federated Asynchronous SGD Robust to Stragglers
Samuel Erickson, Mikael Johansson
In modern machine learning, parallelization of training is an important strategy for increasing scale. Asynchronous stochastic gradient descent (ASGD), which maximizes the utilizat…
cs.LG2026
Decentralized SGD with Controlled Disagreement Finds Flatter Minima
Zesen Wang, Mikael Johansson
Decentralized training is often regarded as inferior to centralized training because the consensus errors between workers are thought to undermine convergence and generalization. T…