Classification with missing and uncertain inputs

Some Bayesian techniques for extracting class probabilities given only partial or noisy inputs are discussed. The optimal solution involves integrating over the missing dimensions weighted by the local probability densities. It is shown how to obtain closed-form approximations to the Bayesian solution using Gaussian basis function networks. Simulation results on the complex task to 3-D hand gesture recognition validate the theory. Using the Bayesian technique, significant information can be extracted, even in the presence of a large amount of noise. The results also show that a classifier that works well with perfect inputs is not necessarily very good at dealing with missing or noisy inputs.<
>