Scalable Robustness
Thomas B. Jones, David H. Ackley · 2016
Complex computational systems can experience undetected faults that produce incorrect outputs. However, error measures can be adopted to quantify these incorrect results and evaluate computational robustness. This paper offers an approach to assessing the worst case scalable robustness (WCSR) of an algorithm paired with an error measure, as well as the i.i.d. average case scalable robustness (ACSRiid). In a case study on four linearithmic and quadratic pairwise sorting algorithms suffering faulty comparisons, we confirm that algorithm efficiency is inversely correlated with algorithm robustness, and more unexpectedly, that only round robin sort -- a quadratic algorithm rarely used in computing -- achieves both ACSRiid and WCSR.