Fast Gumbel-Max Sketch and Its Applications
Yuanming Zhang,Pinghui Wang,Yiyan Qi,Kuankuan Cheng,Junzhou Zhao,Guangjian Tian,Xiaohong Guan
DOI: https://doi.org/10.1109/tkde.2023.3237857
IF: 9.235
2023-01-01
IEEE Transactions on Knowledge and Data Engineering
Abstract:The well-known Gumbel-Max Trick for sampling elements from a categorical distribution (or more generally a non negative vector) and its variants have been widely used in areas such as machine learning and information retrieval. To sample a random element i in proportion to its positive weight v(i), the Gumbel-Max Trick first computes a Gumbel random variable g(i) for each positive weight element i, and then samples the element i with the largest value of g(i) + ln v(i). Recently, applications including similarity estimation and weighted cardinality estimation require to generate k independent Gumbel-Max variables from high dimensional vectors. However, it is computationally expensive for a large k (e.g., hundreds or even thousands) when using the traditional Gumbel-Max Trick. To solve this problem, we propose a novel algorithm, FastGM, which reduces the time complexity from O(kn(+)) to O(k ln k + n(+)), where n(+) is the number of positive elements in the vector of interest. FastGM stops the procedure of Gumbel random variables computing for many elements, especially for those with small weights. We perform experiments on a variety of real-world datasets and the experimental results demonstrate that FastGM is orders of magnitude faster than state-of-the-art methods without sacrificing accuracy or incurring additional expenses.