A parallel randomized neural network on in-memory cluster computing for big data
Tongwu Dai, Kenli Li, Cen Chen · 2017
Owing to its good generalization performance and rapid learning speed, RNN (Randomized Neural Network) is widely used in text classification, image recognition, bioinformatics and other fields. However, the amount of data generated by applications in the real world is growing, and the traditional centralized randomized neural network (RNN) is no longer able to quickly and efficiently learn data that is so large. Therefore, this paper makes use of the distributed framework provided by Flink to realize the randomized neural network (RNN) parallelization algorithm to make up for the deficiency of traditional centralized randomized neural network (RNN) algorithm in dealing with large datasets. By analyzing the characteristics of the traditional centralized randomized neural network (RNN), we find that the most time consuming part is the matrix multiplication operation of the Moore generalized inverse matrix when calculating the output weight vector. Since the matrix multiplication can be decomposed, the distributed matrix can be multiplied by the distributed framework provided by Flink, and then the corresponding output weight vector is centrally calculated. Extensive experiments have validated the effectiveness of our proposed algorithm.