Enhancing robustness and time efficiency of random vector functional link with optimized affine parameters in activation functions and orthogonalization

Published: 01 Jan 2024, Last Modified: 25 Jan 2025Appl. Soft Comput. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Proposes a robust & time-efficient affine transformation-based random vector functional link approach, called ‘RT-ATRVFL’.•It calculates affine parameters based on sampling of only two column vectors from the entire hidden layer output.•It increases the entropy and avoids the concentration of hidden node output in the saturated region of activation function.•It can eradicate the vanishing gradient problem. It gives better accuracy, robustness, and reduces the computational cost.•It exploits cloglogm activation function and orthogonalization for enhanced robustness, and reduction in time complexity.
Loading