Negative Selection in Negative Correlation Learning

Negative correlation learning is an ensemble learning approach that is able to create negatively correlated learners simultaneously and cooperatively in a committee machine. One problem in negative correlation learning is that the learning error functions are defined in the same way for all individual learners. Learners have little choice in making their own decisions on how to learn a given data. Two different negative selections have been introduced in negative correlation learning for letting individual learners be able to adapte the learning error functions in the whole learning process. The first negative selection is based on the opposition learning which some learners in a committee could turn to learn the opposite targets rather than the correct targets. The second negative selection is through difference learning in which each learner could decide to weaken or strengthen its learning signal on each data based on how different it is to the rest of learners in the committee machine. It is expected that such negative selections would well deal with the trade off between the training accuracy by the learners and the diversity among the learners in the committee machine. Experimental results were given to compare the learning behaviors in the committee machines by negative correlation learning with the two different negative selections.