Machine Learning Aided Network Slicing

Future 5G wireless networks will need to deal with increasing heterogeneity in offered services, performance requirements, and communication technologies. Network slicing has appeared as a technique to deal with the complex management of these future networks. Particularly, in wireless networks, link bitrate depends on several factors (e.g., interference). In a wire-wireless network scenario, any mechanism that proposes to manage and allocate resources to implement slicing in wireless devices will require knowledge of the channel characteristics. In this paper, we present a Deep Learning algorithm to predict if a service provider will be able to fulfil a new network slice request given the conditions of the channel and the allocated resources. As part of the Deep Learning algorithm architecture, given the sequentially of the transmission data, we designed and implemented a Long Short Memory Term network for predicting the channel conditions in the near future. Results show that the propose Deep Learning algorithm reduce the number of false positive allocations by a 75%.