SEM-O-RAN is a network slicing framework for next-generation mobile networks capable of orchestrating the offloading of Machine Learning tasks at the Edge. The framework uses the interfaces and architecture proposed by the O-RAN Alliance to allocate network and computing resources, maximizing the number of tasks that can be processed at the Edge, depending on the semantics of the tasks and end-to-end precision and latency requirements -end.

The invention is a framework dedicated to slicing on mobile networks. The objective of the framework is to admit and allocate resources for the execution of tasks, delegated to Edge processing by devices with a limited budget of computing resources for local execution. Given a set of tasks submitted for offloading to the edge, the system selects a subset of the submitted tasks and guarantees their execution within the specified requirements thanks to the allocation of the appropriate computing and network resources. The invention consists of two logical components: the Semantic Edge Slicing Module (SESM), dedicated to the allocation of resources for the various tasks, and the Semantic Deep Learning Analyzer (SDLA) which calculates, for each task, the level of precision obtainable on the task results and what end-to-end latency would be encountered for each possible resource allocation and compression allowed.

• Edge computing
• Machine Learning
• Mobile networks
• Computer vision

• Semantic-based edge computing task optimization
• Greater number of tasks managed (up to 53% more with the same resources)
• Reduction of CAPEX and OPEX
• Easily integrated with O-RAN platforms