Automatic Design Space Exploration of Approximate Algorithms for Big Data Applications

摘要: Nowadays, Big Data represents a new era in data exploration, manipulation and utilization as it requires an effective rethinking of platforms, computing paradigms, architectures algorithms. Indeed, performance constraints cannot be sustained by traditionally designed platforms design methodologies need to defined. Approximate is promising strategy, able deal with these issues, trades off accuracy parameters, such energy, time required hardware resources. It allows designers exploit the inherent resiliency algorithms, i.e. when application tolerates approximate result, develop platform lower power consumption, smaller area footprints elaboration time. In this paper, we present exploration methodology whereby generic described means C/C++, are implemented automatically tuning respect given maximum loss, illustrating preliminary experimental result on some