Fig. 2

Performance of the exponentiation protocol. The data points are based on an average of 100 runs per datapoint. We observe a linear scaling in the size of the vector x. Remark: we need \(n(n-1)/2\) invocations of the exponentiation, where n is the number of sample (or patients), resulting in quadratic scaling in the number of samples. A more elaborate explanation of the legend; Green: exponents are assumed to be in the interval [0, 12]. No truncation is performed to enforce this, resulting in zero secure comparisons to perform the exponentiation; Orange: exponents are assumed to be in the interval \([-12, 12]\). No truncation is performed to enforce this, however one secure comparison needed to deal with negative exponents; Blue: given an interval \([-x, y]\) (e.g., \([-12, 12]\)), all exponents are truncated to fit in this range to prevent overflows. Two secure comparisons are needed to achieve this