Multi-Objective Application-Driven Approximate Design Method

Approximate Computing (AxC) paradigm aims at designing computing systems that can satisfy the rising performance demands and improve energy efficiency. AxC exploits gap between level of accuracy required by users, actual precision provided system, for achieving diverse optimizations. Various techniques have been proposed so far in literature different abstraction levels from hardware to software. These successfully utilized combined realize approximate implementations applications various domains (e.g. data analytic, scientific computing, multimedia signal processing, machine learning). Unfortunately, state-of-the-art approximation methodologies focus on a single level, such as combining elementary components (e.g., arithmetic operations) which are firstly approximated using component-level metrics then provide good trade-off efficiency application level. This hinders possibility designers explore opportunities, optimized implementation targets. Therefore, we designed implemented E- \mathbb IDEA, an automatic framework provides application-driven approach find best versions given targeting (i.e., software). IDEA compounds 1) source-to-source manipulation tool 2) evolutionary search engine automatically variants perform Design-Space Exploration (DSE). The latter results set non-dominate solutions terms Experimental validate effectiveness flexibility generating applications, accuracy/error