Communication-efficient sparse regression: a one-shot approach
This work addresses communication efficiency in distributed machine learning for sparse regression, but it is incremental as it builds on existing debiased lasso methods.
The authors tackled distributed sparse regression in high-dimensional settings by proposing a one-shot approach that averages debiased lasso estimators, achieving convergence rates comparable to the lasso when data is not overly split across machines, and extended it to generalized linear models.
We devise a one-shot approach to distributed sparse regression in the high-dimensional setting. The key idea is to average "debiased" or "desparsified" lasso estimators. We show the approach converges at the same rate as the lasso as long as the dataset is not split across too many machines. We also extend the approach to generalized linear models.