ELMs are just neural nets with random hidden weights. This idea has been around since the 60's before being rebranded to get unoriginal papers published. This

ELMs are just neural nets with random hidden weights. This idea has been around since the 60's before being rebranded to get unoriginal papers published. This is equivalent to a low-rank Fourier approximation of Gaussian process regression (also known as random features) for some kernel. Random kitchen sink approximations are similar, but better performing since the weight value distribution actually has theoretical justification. RKS is a really great way to use approximate Gaussian process regression on larger datasets.