Current version only has the recursive Gaussian projection step by moment matching when the input distribution is Gaussian.
An alternative is linearising the GP at the mean of this Gaussian using Taylor expansion, and propagating this Gaussian thru a linear function.
The output distribution is also Gaussian, which can be used to propagate forward.
Current version only has the recursive Gaussian projection step by moment matching when the input distribution is Gaussian. An alternative is linearising the GP at the mean of this Gaussian using Taylor expansion, and propagating this Gaussian thru a linear function. The output distribution is also Gaussian, which can be used to propagate forward.
Also need to compare with moment matching.