Approximate Computing: Challenges and Opportunities

Approximate computing is gaining traction as a computing paradigm for data analytics and cognitive applications. In this paper, we demonstrate that multiple approximation techniques can be applied to applications in these... [ view full abstract ]

Approximate computing is gaining traction as a computing paradigm for data analytics and cognitive applications. In this paper, we demonstrate that multiple approximation techniques can be applied to applications in these domains and can be further combined together to compound their benefits. In assessing the potential of approximation in these applications, we took the liberty of changing multiple layers of the system stack: architecture, programming model, and algorithms. Across a set of applications from DSP, robotics, and machine learning, our results show that without incurring loss of quality, hot loops can be perforated by an average of 50%, the width of the data used in the computation can be reduced to 10-16 bits, and synchronization can be relaxed in parallel applications to achieve significant performance/energy benefits. Furthermore, Finally, the benefits compounded when these techniques are applied concurrently. Our results affirm the wide applicability of approximate computing with potential for compounded benefits from applying multiple techniques. To exploit these benefits it is essential to re-think multiple layers of the system stack to embrace approximations ground-up, thereby moving the applications into a world in which the architecture, programming model, and even the algorithms used are all fundamentally designed for approximate computing.