Data Set Selection, 献给还在苦海挣扎的人们

刚才无意中看到一篇好文，转载如下，希望能为那些依然在痛并研究着的同志们减轻一点痛苦

http://www.jmlg.org/papers/laloudouana03.pdf

Abstract

We introduce the community to a new construction principle whose practical implications are very broad. Central to this research is the idea of improving the presentation of algorithms in the literature and making them more appealing. We define a new notion of capacity for data sets and derive a methodology for selecting from them. Our experiments demonstrate that even not-so-good algorithms can be shown signifficantly better than competitors. We present some experimental results, which are very promising.