Some Nonlinear Inference Problems

In this chapter we discuss several non-linear inference problems. One issue that we shall very briefly discuss in the beginning is direction estimation, which is for example relevant to scene and motion reconstruction in computer vision. A geometric formulation points to a generalization to spherical manifolds of the familiar Gaussian inference on linear-spaces. Next we discuss in some detail non-linear support vector machines which is a very important subject with many potential applications. Finally we present a critical view of Neural Networks a widespread non-linear inference tool which seems to have become the exclusive basic subject of statistical learning.