Incremental machine learning theorem and algorithm based on DSM method

An incremental and efficient algorithm is the key to knowledge acquisition and machine learning. The DSM is defined in the paper, the important principle of the best knowledge reduction is found and a new method is put forward through analyzing the elements m/sub u//sup d/ & m/sub u//sup s/ in DSM. The reduction efficiency is improved and the reduced rules are the least and the number of the attributes in the rules is the least too, so it is the best knowledge reduction. We also put forward a new incremental learning method based on DSM. The method can automatically maintain the rules when the new instances are added.