This paper presents a new method for continous learning based on data transformation. The proposed approach is applicable where individual training datasets are separated and not sharable. This approach includes a long short term memory network combined with a pooling process.