Towards Generic and Scalable Word-Length Optimization

摘要: In this paper, we propose a method to improve the scalability of Word-Length Optimization (WLO) for large applications that use complex quality metrics such as Structural Similarity (SSIM). The input application is decomposed into smaller kernels avoid uncontrolled explosion exploration time, which known noise budgeting. main challenge addressed in paper how allocate budgets each kernel. This requires capturing interactions across kernels. idea characterize impact approximating kernel on accuracy/cost through simulation and regression. Our approach improves while finding better solutions Image Signal Processor pipeline.