I am reading about relevance vector machines as a potential methodology to address a regression problem with prediction of a continuous variable, where I more specifically want to work with a model which is suitable for quantifying uncertainty.
In this article they compare RVMs with SVMs, and discuss that unlike SVMs one can attain the associated uncertainty of a prediction when using RVMs. However they also mention that RVMs can get stuck in local optima rather than finding the global optima which SVMs do. How RVMs work is also addressed in How does a Relevance Vector Machine (RVM) work? but from an approach of quantifying uncertainty, is it possible to evaluate an estimate of RVMs, and it's associated uncertainty if we can't ensure it's a global optima?
