Parallelized extreme learning machine for online data classification

The challenges raised by the massive data are being managed by the community through the advancements of infrastructure and algorithms, and now the processing of fast data is becoming a new hurdle to the researchers. Extreme Learning Machine (ELM) is a single-layer learning model with reliable performances and it is computationally simpler than the new generation deep architectures. ELM process the data in batches and the model has to be rerun while updates happening in the datasets. In the theoretical background of ELM, the past knowledge cannot be reused for improving the performance in online learning where the data set will be updated with mini-batches. In this paper, we have introduced a knowledge base to deal with the remembrance of knowledge in ELM. The architecture of the proposed model is designed to process mini-batches of any size to speed up the processing of the data on its arrival. A group of data sets with different properties such as sparse and feature dimensions is used in the experiments to evaluate our method. The performance of the algorithm is compared with a set of benchmarked classifiers and stream classifiers in the scikit-learn public platform. It is observed that our method could perform better in most of the experiments. It clear in the results that the Parallel ELM model outperformed the other methods in the training time across all the datasets. The consistent performance of our method shows the significance of parallel algorithms of ELM that can remember past knowledge.