Ensemble Decomposition Learning for Optimal Utilization of Implicitly Encoded Knowledge in Biomedical Applications

Olga V. Senyukova and Valeriy V. Gavrishchaka


ensemble decomposition learning, boosting, single-example learning


Ensemble learning techniques, especially boosting-like, provide practical and efficient means for improving performance of existing domain-specific base classifier models. This is achieved by combination of complementary models that are experts in different regimes of the considered complex system. Therefore, partial information of many different regimes becomes implicitly encoded in the obtained ensemble. However, only aggregated output is used in the majority of applications, while the rich internal structure of the ensemble is completely ignored. Extraction of this underutilized knowledge could be formalized in terms of ensemble decomposition learning (EDL) techniques. We outline one of such frameworks based on existing single-example learning (SEL) algorithms. The experiments on real physiological data showed that the proposed approach is suitable for classification of rare diseases or complex physiological cases with one or few training samples.

Important Links:

Go Back