Distributed Edge Computing Combined with Reinforcement Learning for Low-Latency Probabilistic Skyline Queries

Probabilistic skyline refers to a set of datasets that satisfy multiple conditions in a dataset with uncertainty. Unlike traditional skyline queries, the probabilistic skyline considers the uncertainty of data items and calculates the probability of each data becoming part of the skyline. In this article, we adopt a reinforcement learning approach to dynamically adjust the threshold values in the distributed edge computing architecture, which is different from the traditional method where each edge server uses a preset fixed threshold value for data filtering. This allows the model to dynamically adjust the threshold values according to the characteristics of the data flow and effectively control the size of the probabilistic skyline search set for each server, thereby significantly reducing unnecessary data transmission costs. This flexible adjustment mechanism reduces the waste of network resources and effectively improves system performance, thereby significantly reducing the latency (or computation time) for obtaining probability skyline sets by more than 50%.