Improved algorithms for distributed boosting
Jeff Cooper, Lev Reyzin · 2017
We introduce two distributed boosting algorithms. Our first algorithm uses the entire dataset to train a classifier and requires a significant amount of communication among the distributed sites. Our second algorithm requires very little communication but uses a subsample of the dataset to train the final classifier. Both of our algorithms improve upon existing practical distributed boosting algorithms. Further, both are competitive with AdaBoost when it is run with the entire dataset.