Cross-class modeling for heterogeneous network-slice requests

Develop a reward and constraint surrogate model that exploits similarities between URLLC and eMBB request classes rather than maintaining separate exploration estimates and training datasets for each class.

Background

In the heterogeneous-traffic experiments, the proposed method maintains separate reward and constraint exploration estimates for URLLC and eMBB requests to avoid pooling observations with substantially different traffic loads and SLA parameters. This class-specific treatment means that each model is trained on fewer observations and cannot exploit potential similarities between the two request classes. The paper explicitly identifies development of a model that shares such information as future work.