Quantization Error and Accuracy-Performance Tradeoffs for Embedded Data Mining Workloads

Abstract

Abstract. Data mining is the process of automatically finding implicit, previously unknown and potentially useful information from large volumes of data. Embedded systems are increasingly used for sophisticated data mining algorithms to make intelligent decisions while storing and analyzing data. Since data mining applications are designed and implemented considering the resources available on a conventional computing platform, their performance degrades when executed on an embedded system. In this paper, we analyze the bottlenecks faced in implementing these algorithms in an embedded environment and explore their portability to the embedded systems domain. Particularly, we analyze the floating point computation in these applications and convert them into fixed point operations. Our results reveal that the execution time of five representative applications can be reduced by as much as 11.5× and 5.2× on average, without a significant impact on accuracy.

Dynamic knobs for responsive power-aware computing

“…Trading accuracy of computation for other benefits is a wellknown technique. It has been shown that one can trade off accuracy for performance [17,32,40,43,44], energy consumption [13,14,17,21,34,43,47] and fault tolerance [14,43,47]. The Dynamic Knobs system presented in this paper, along with loop perforation [27,39] and task skipping [43,44], is unique in that it enables applications to adaptively trade accuracy for performance and does so without requiring a developer to change the application source code.…”

Dynamic knobs for responsive power-aware computing

“…It has been shown that one can trade off accuracy for performance [16,32,40,43,44], energy consumption [12,13,16,20,34,43,47] and fault tolerance [13,43,47]. The Dynamic Knobs system presented in this paper, along with loop perforation [26,39] and task skipping [43,44], is unique in that it enables applications to adaptively trade accuracy for performance and does so without requiring a developer to change the application source code.…”

On the path to sustainable, scalable, and energy-efficient data analytics: Challenges, promises, and future directions

“…Algorithms implemented using fixed-point arithmetic [11] trade precision for energy-and-computational efficiency, replacing expensive floating-point operations with a sequence of faster integer operations. Compute-intensive parts of kernels described in Table I depend heavily on of resources on accelators like FPGAs, making them prime candidates for fixed-point optimization.…”