ELM ∗ : distributed extreme learning machine with MapReduce

Elastic extreme learning machine for big data classification

“…Thus, we can use MapReduce framework to speedup the computation of output weight vector β. The process of calculating matrices U and V based on MapReduce framework is shown in Algorithm 1 [28]. In the Initialize method of Mapper, we initialize two arrays, u and v, which are used to store the intermediate summation of the elements in matrices U and V respectively.…”

“…As a variant of ELM, distributed ELM (i.e. PELM [27] and ELM n [28]) based on MapReduce [29][30][31] can resolve the V1 (Volume) problem of Big Data. However, it is quite common in big data classifications that some new training data arrived, some old training data expired and some error training data corrected.…”

“…A simple and direct way is to retraining the ELM n [28] using the whole training dataset. Obviously, this kind of method is timeconsuming, since the training dataset is very large.…”

“…Obviously, this kind of method is timeconsuming, since the training dataset is very large. Therefore, it is essential to improve the ELM n [28] algorithm, in order to support the functionalities such as incremental learning, decremental learning, and correctional learning. There is a great overlap between the old training dataset and the new one, if the overlapped information can be used, the retraining cost will be reduced greatly.…”

“…There is a great overlap between the old training dataset and the new one, if the overlapped information can be used, the retraining cost will be reduced greatly. Therefore, in this paper, an Elastic ELM (E 2 LM) is proposed to improve ELM n [28] with the abilities of incremental learning, decremental learning, and correctional learning. The contributions of this paper can be summarized as follows.…”