Optimizing the Capacity of Extreme Learning Machines for Biomedical Informatics Applications

Abstract

the paper discusses a way of growing the capability of severe deep learning machines (ELM) for biomedical informatics programs. This technique involves varying the dimensions of the training set, the variety of neurons in the hidden layer, the kind of activation function used, the range of different sorts of neurons used and the regularization techniques used.so that it will maximize the ability of ELM fashions for biomedical informatics applications, the authors advocate using a stepwise seek technique to determine the fine combination of parameters. This technique entails first defining a baseline or initial configuration of ELM parameters and then iteratively enhancing that configuration via deep learning from the empirical effects of each optimization step. distinctive regularization strategies are carried out at every optimization step which could involve pruning hidden neurons as a way to reduce the complexity of the model. The authors verified their technique on publicly to be had datasets consisting of the Michigan Imputation of missing Values (MIMV) dataset and on a dataset of clinical records from the branch of Anesthesiology, branch of health and Human services.