Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Tristique et egestas quis ipsum suspendisse. Quisque id diam vel quam elementum. Risus viverra adipiscing at in tellus integer feugiat scelerisque. Purus in massa tempor nec feugiat nisl pretium.
Quisque id diam vel quam elementum. Risus viverra adipiscing at in tellus integer feugiat scelerisque. Purus in massa tempor nec feugiat nisl pretium. Morbi blandit cursus risus at ultrices mi tempus imperdiet nulla. Porta lorem mollis aliquam ut porttitor.
In order to improve communication efficiency and adversarial robustness, the well known XGBoost libraries have been modified to execute in a federated setting, thus allowing the training of multiple number of trees using XGBoost functionalities along with associated local data set(s) before passing the trees to a centralised server for aggregation purposes.
In a conventional approach, client devices train just one single tree and send that single tree to the server for aggregation. The inventive approach is referred to as “tree plus” in that client devices can train trees locally and pass a number of trees to a centralised service or server, significantly improving processing performance and reducing power costs on the devices. In some embodiments, normalization of one or more parameters, such as a “learning rate” allows different client devices to process a different number of datapoints.
