Advancing synthesis of decision tree-based multiple classifier systems: an approximate computing case study

摘要: So far, multiple classifier systems have been increasingly designed to take advantage of hardware features, such as high parallelism and computational power. Indeed, compared software implementations, accelerators guarantee higher throughput lower latency. Although the combination classifiers leads classification accuracy, required area overhead makes design a accelerator unfeasible, hindering adoption commercial configurable devices. For this reason, in paper, we exploit approximate computing paradigm trade off for accuracy. In particular, starting from trained DT models employing precision-scaling technique, explore decision tree variants by means objective optimization problem, demonstrating significant performance improvement targeting field-programmable gate array