Der Einsatz von Gaußprozessen zur Beschleunigung der automatischen Wissensakquisition

Knowledge acquisition is the bottleneck in the process of developing and maintaining knowledge based systems and can be performed in a direct, indirect or automatic manner. While direct and indirect methods depend on the presence of experts and knowledge engineers, the automatic approaches depend on data e.g. from parameter variation studies. The data is later used in a KDD-process (knowledge discovery in databases) that can train useful prediction models. Here a conflict arises: More data will increase the reliability of the prediction model but will also increase the costs and the time necessary for the variation study. This paper reports about the utilization of gaussian process based machine learning that can handle the described conflict by enabling an adaptive sampling of training data.