Applied science

About: Applied science is a research topic. Over the lifetime, 1178 publications have been published within this topic receiving 19920 citations. The topic is also known as: applied sciences.

Approximate computing: An integrated cross-layer framework

Abstract: Venkataramani, Swagath PhD, Purdue University, December 2016. Approximate Computing: An Integrated Cross-layer Framework. Major Professor: Anand Raghunathan. We have witnessed a fundamental shift in the nature of workloads executed by computing platforms across the spectrum, from mobile and deeply-embedded devices to servers and data centers. Increasingly, computing platforms need to analyze, organize and search through large amounts of real-world data, intelligently interact with the physical world, be context-aware, and present more natural human interfaces. These tasks do not involve the computation of a golden answer or unique numerical result. Instead, they need to produce outputs that are good-enough or of sufficient quality. Such workloads possess intrinsic application resilience, or the ability to produce outputs of acceptable quality even when a large fraction of their computations are performed in an imprecise or approximate manner. Intrinsic application resilience offers an entirely new dimension along which computing platforms can be optimized. However, the design of computing platforms still continues to be guided by the dogma that every computation must be executed with the same strict notion of correctness. With the demand for computing performance growing unabated on the one hand, while traditional benefits due to technology scaling diminish on the other, it is important to leverage this new source of efficiency. A new design approach, called approximate computing (AxC), leverages the flexibility provided by intrinsic application resilience to realize hardware or software implementations that are more efficient in energy or performance. Approximate computing techniques forsake exact (numerical or Boolean) equivalence in the execution of some of the application’s computations, while ensuring that the output quality is