2019/03/07 by Sajad Mehrizi, Mehrizi, Sajad, Anestis Tsakmalis +5
Computer Science · Social Sciences · #Caching and Content Delivery #FOS: Electrical engineering #Human Mobility and Location-Based Analysis #Recommender Systems and Techniques #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1903.03065
openalex publication_date 2019/03/07 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Edge-caching is recognized as an efficient technique for future cellular\nnetworks to improve network capacity and user-perceived quality of experience.\nTo enhance the performance of caching systems, designing an accurate content\nrequest prediction algorithm plays an important role. In this paper, we develop\na flexible model, a Poisson regressor based on a Gaussian process, for the\ncontent request distribution.\n The first important advantage of the proposed model is that it encourages the\nalready existing or seen contents with similar features to be correlated in the\nfeature space and therefore it acts as a regularizer for the estimation.\nSecond, it allows to predict the popularities of newly-added or unseen contents\nwhose statistical data is not available in advance. In order to learn the model\nparameters, which yield the Poisson arrival rates or alternatively the content\n\popularities, we invoke the Bayesian approach which is robust against\nover-fitting.\n However, the resulting posterior distribution is analytically intractable to\ncompute. To tackle this, we apply a Markov Chain Monte Carlo (MCMC) method to\napproximate this distribution which is also asymptotically exact. Nevertheless,\nthe MCMC is computationally demanding especially when the number of contents is\nlarge. Thus, we employ the Variational Bayes (VB) method as an alternative low\ncomplexity solution. More specifically, the VB method addresses the\napproximation of the posterior distribution through an optimization problem.\nSubsequently, we present a fast block-coordinate descent algorithm to solve\nthis optimization problem. Finally, extensive simulation results both on\nsynthetic and real-world datasets are provided to show the accuracy of our\nprediction algorithm and the cache hit ratio (CHR) gain compared to existing\nmethods from the literature.\n