A new SVM bound for information retrieval

Support vector machines have met with a significant success in machine learning field, especially handling high dimensional classification tasks. Although various performance estimators for SVM have been proposed, the based bound of kernel methods have quite margin to improve the capability. In this paper, a novel Alpha-SV bound has been proposed, which are based on the span bound of the leave-one-out procedure. Experiments have shown that the proposed estimators are both effective and stable.