ELM and ELM_2: Modern Extreme Learning Machines with Source Code Implementation

ELM (Extreme Learning Machine) and ELM_2 represent modern, high-performance learning machines gaining significant popularity in machine learning research. The EELM source code implements an efficient ELM variant that was published in Neural Computing journal by researcher Yuguang Wang, featuring optimized hidden layer node selection and faster matrix computation methods. Meanwhile, YELM provides an improved ELM implementation published by Yuboyuan, incorporating enhanced regularization techniques and refined weight initialization strategies. This software package performs comparative analysis of these three methodologies (ELM, EELM, YELM) to evaluate their respective advantages and limitations in terms of classification accuracy, training speed, and generalization capability. The core implementation includes modular functions for data loading, feature normalization, random weight generation, and output weight calculation using Moore-Penrose pseudoinverse. For detailed theoretical foundations and mathematical derivations, users should consult the original publications by the respective authors. To execute the comparison framework, users need only run the main entry point pendigits_mean.m script, which automatically handles dataset loading (using the pendigits benchmark dataset), parameter initialization, cross-validation setup, and generates comprehensive performance metrics including training time, testing accuracy, and standard deviation results across multiple runs. The code structure follows MATLAB best practices with clear function separation and detailed inline comments for easy modification and extension.