An outlier detection method and its application to multicore-chip power estimation

摘要

Machine learning has been recently applied to solve challenging research problems in the EDA area. The performance of the machine learning algorithms are vulnerable to effects such as the outliers and the insignificant features in the input training data set. In this paper, we propose a model averaging method, together with an outlier detection method, to make the machine learning process more robust to such contaminating effects. Results on artificial and chip power estimation data sets show that our method behave much better than the conventional ordinary least square method which is widely used in the machine learning community.