I also liked the ideas presented in the paper although i think there were too many. To me it seemed that there was not a single idea on which the paper was focusing.

Feature selection in itself is a very important and interesting task. I think that the paper showed one particular way of performing feature selection.

I think the general result that "feature selection methods can be applied to classification problems, [where there are only a few data points and those data points are high dimensional]" is not surprising at all.

Also, i found the comparison to classification results using a random subset of features somewhat unfair. I think it would be interesting so see how classifiers such as SVMs would perform on the high dimensional data set, or on the dataset, which is obtained after applying traditional dimensionality reduction techniques.