Supervised Potentiality Actualization Learning for Improving Generalization Performance

The present paper proposes an application of potentiality learning to supervised learning. The potentiality has been developed as a measure of the importance of components in the self-organizing maps (SOM) to extract important input neurons. The main characteristics lies in its simplicity and thus it can be easily implemented. If it is possible to use it for conventional supervised learning, better performance can be expected with much simpler computational method. The potentiality is defined by the variance of input neurons and it is incorporated into supervised learning. Using the potentiality inside, two data sets were used to evaluate the performance. The results show that the potentiality method outperformed ones without it and other conventional methods in terms of generalization performance.