An Ensemble Approach for Data Fusion with Learn++

We have recently introduced Learn++ as an incremental learning algorithm capable of learning additional data that may later become available. The strength of Learn++ lies with its ability to learn new data without forgetting previously acquired knowledge and without requiring access to any of the previously seen data, even when the new data introduce new classes. Learn++, inspired in part by AdaBoost, achieves incremental learning through generating an ensemble of classifiers for each new dataset that becomes available and then combining them through weighted majority voting with a distribution update rule modified for incremental learning of new classes. We have recently discovered that Learn++ also provides a practical and a general purpose approach for multisensor and/or multimodality data fusion. In this paper, we present Learn++ as an addition to the new breed of classifier fusion algorithms, along with preliminary results obtained on two real-world data fusion applications.