Showing 1–1 of 1 results for author: Fedchenko, V

Feedforward Neural Networks for Caching: Enough or Too Much?

Authors: Vladyslav Fedchenko, Giovanni Neglia, Bruno Ribeiro

Abstract: We propose a caching policy that uses a feedforward neural network (FNN) to predict content popularity. Our scheme outperforms popular eviction policies like LRU or ARC, but also a new policy relying on the more complex recurrent neural networks. At the same time, replacing the FNN predictor with a naive linear estimator does not degrade caching performance significantly, questioning then the role… ▽ More We propose a caching policy that uses a feedforward neural network (FNN) to predict content popularity. Our scheme outperforms popular eviction policies like LRU or ARC, but also a new policy relying on the more complex recurrent neural networks. At the same time, replacing the FNN predictor with a naive linear estimator does not degrade caching performance significantly, questioning then the role of neural networks for these applications. △ Less

Submitted 16 October, 2018; originally announced October 2018.