Abstract

Extreme Learning Machines (ELM) are feedforward neural networks where most of the connections are randomly fixed, and only the output weights are trained [1]. On small scales, ELMs perform comparably to fully trained networks. Due to their low complexity, ELMs are particularly interesting in scenarios where compact, low-cost and non-electronic alternatives to the von-Neumann architecture are desirable, like in edge-computing [2].

© 2023 IEEE