On the Energy Efficiency of Machine Learning Frameworks
With the emphasis on machine and deep learning infrastructures, there is a concerted focus in the industry in the areas ofhardware (e.g. learning accelerators) and software (e.g. deep learning models, open-source frameworks and supporting libraries). The focus of these efforts has been on improving the accuracy and performance of the learning algorithms. However, the high speed and accuracy are at the cost of energy consumption. While the size of data sets grows exponentially, the energy demand for runningwith such data sets increases rapidly. It is highly desirable to design learning frameworks and algorithms that are both accurate and energy efficient. In order to develop such energy efficient software frameworks we would need a way to estimate and predict the energy cost of these.In this presentation we will discuss an analysis setup, and results of our study comparing the energy efficiency and power consumption behavior of learning frameworks on hardware platforms. The benefit would be to provide a detailed analysis of machine learning workloads, which can subsequently set the stage to facilitate the design of energy efficient learning solutions.