I'm going to assume you have done proper pre-processing (background correction, normalisation and summarisation) using RMA or likewise.

For translating probe ids to genes you might want to use a custom CDF instead of the supplied one (http://brainarray.mbni.med.umich.edu/brainarray/Database/CustomCDF/cdfreadme.htm).

For a naive approach, you could just use the median values of the genes. Most people discourage this, though, since multiple expression levels for a gene may come from different isoforms.