Joint boosting is an ensemble machine learning technique for multi-class classification and multi-task learning that trains multiple classifiers simultaneously by sharing features across different classes. Instead of learning independent binary classifiers for each individual category, the algorithm iteratively selects weak learners and underlying features that can be shared among subsets of classes to minimize joint training error. This shared feature selection substantially reduces both computational and sample complexity, enabling models to scale efficiently as the number of target categories grows while encouraging the learning of generic, reusable representations that generalize effectively across tasks.
