LSTM based content prediction for edge caching using federated learning approach

Abstract

With rapid expansion and worldwide penetration of internet usage, there has been a rapid growth and development in the field of communication technology. To meet a never ending demand of excellence in quality and computation, a relatively new and effective computation theory called Edge computing is making its mark. Edge computing basically means the computing which is done at or near the data source instead of relying on the cloud to do all the work which enhances network performance by reducing latency. With Edge computing and Edge caching we seek to integrate federated learning approach by training the model across multiple edge nodes that have thier own local environment, without exchanging them which will eventually turn into Edge Intelligence by increasing system level optimization making content delivery faster than before. In a whole in this research topic we aim to investigate service provisioning in edge computing which will make our daily used devices more efficient in terms of performance and keep our personal data secured with the help of federated learning approach. Accurate content prediction combined with optimized caching promises to be a future-proof solution. We adopt a hierarchy based three layer system architecture in which we integrate federated learning with LSTM for predicting content based on view count. With our FedPredict algorithm we intend to maximize cache hit so that the network flow remains optimized. Lastly, we look into potential optimization our algorithm and address some areas of improvement regarding distributed learning systems.