Efficient Online Random Sampling via Randomness Recycling

Abstract: ``Randomness recycling'' is a powerful algorithmic technique for reusing a fraction of the random information consumed by a randomized algorithm to reduce its entropy requirements. This article presents a family of efficient randomness recycling algorithms for sampling a sequence $X_1, X_2, X_3, \dots$ of discrete random variables whose joint distribution follows an arbitrary stochastic process. We develop randomness recycling strategies to reduce the entropy cost of a variety of prominent sampling algorithms, which include uniform sampling, inverse transform sampling, lookup table sampling, alias sampling, and discrete distribution generating (DDG) tree sampling. Our method achieves an expected amortized entropy cost of $H(X_1,\dots,X_k)/k + \varepsilon$ input random bits per output sample using $O(\log(1/\varepsilon))$ space, which is arbitrarily close to the optimal Shannon entropy rate. The combination of space, time, and entropy properties of our method improve upon the Han and Hoshi interval algorithm and Knuth and Yao entropy-optimal algorithm for sampling a discrete random sequence. An empirical evaluation of the algorithm shows that it achieves state-of-the-art runtime performance on the Fisher-Yates shuffle when using a cryptographically secure pseudorandom number generator. Accompanying the manuscript is a performant random sampling library in the C programming language.