Architecture-parameters-guided approximate computing technique

This paper proposes an approximate computing technique for the guidance of architectural characteristic parameters. The technique analyzes the relationship between architectural parameters and approximate sensitive features. The parameters with great influence on approximate computing are selected by machine learning method.The classification and regression tree is used to analyze the approximate sensitivity of application and select the approximate degree according to the analysis results. Experimental results show that the method proposed by the paper can select the appropriate approximate degree for different application, and achieve the tradeoff optimization between computational accuracy and energy efficiency.