Edge/Cloud Slice Resource Allocation for Beyond 5G Networks With Distributed LSTM

Efficient resource allocation among slices/users with different Service Level Agreements (SLAs) is a critical task in 5G+ networks, which has prompted recent research into Deep Neural Networks (DNNs). However, challenges arise when dealing with edge resources, including the ability to rapidly scale resources (in the order of milliseconds), and the cost of transmitting large data volumes to a cloud for centralized DNN-based processing. Addressing these issues, we introduce a novel architecture based on Distributed Deep Neural Networks (DDNN). This architecture features a compact set of DNN layers located at the network's edge, designed to function as an autonomous resource allocation unit. Complementing this, there is an intelligent offloading mechanism that delegates a fraction of hard decisions to additional DNN layers situated in a remote cloud (when needed). To implement offloading, we propose a theoretically informed method that learns to mimic an oracle that knows which sample will benefit from additional processing in the cloud. We compare this to a previously proposed heuristic, based on a Bayesian-confidence mechanism. We investigate the interplay of (offline) joint training of the DDNN exits and the ML-based offloading mechanism, and demonstrate that our architecture resolves more than 50% of decisions at the edge with no additional penalty compared to centralized models as well as consistently outperforms previous methods.