Cost-efficient And Quality Assured Approximate Computing Framework Using Nerual Network

Approximate computing is a promising design paradigm that introduces the error dimension into the original design space.By allowing the inexact computation in error tolerance applications,approximate computing can gain both performance and energy-efficiency.A neural network(NN)is a universal approximator in theory and possesses a high level of parallelism.The emerging DNN accelerators deployed with NN-based approximator is thereby a promising candidate for approximate computing.Nevertheless,the approximation result must satisfy the users' requirement,and it varies across different applications.We normally deploy a NN-based classifier to ensure the approximation quality.Only the inputs predicted to meet the quality requirement can be executed by the approximator.The potential of these two NNs,however,is not fully explored;the involving of two NNs in approximate computing imposes critical optimization questions,such as two NN's distinct views of the input data space,how to train the two correlated NNs,what are their typologies.In this paper,we propose a novel NN-based approximate computing framework with quality control.We advocate a co-training approach that trains the classifier and the approximator alternately to maximize the consistency of the two NNs on the input space.In each iteration,we coordinate the training of the two NNs with a judicious selection of training data.Next,we explore different selection policies and propose to enhance the invocation of the approximate accelerator.Also,we propose a multiclass training algorithm to maximize the invocation of approximators by training those data discarded in the co-training algorithm.Experimental results show significant improvement on energy-efficiency and quality compared to the existing NN-based approximate computing frameworks.