ML Model Selection for Cellular Edge Computing

The patent segments the data fidelity space into multiple distinct fidelity levels (e.g., high, medium, low). For each fidelity level, a separate ML model is trained specifically on data with that fidelity characteristic. This segmentation allows each model to specialize in its designated fidelity range, achieving high accuracy for that specific condition without the complexity burden of a single universal model.

The patent implements local quality by training each ML model with data specifically tailored to its target fidelity level. Model A is trained exclusively on high-fidelity data, Model B on medium-fidelity data, and so on. Each model develops local expertise optimized for its specific fidelity condition, rather than attempting to generalize across all fidelities with a single model.

The patent introduces dynamic model selection based on the detected data fidelity level. The system continuously monitors the fidelity of incoming data and dynamically selects the appropriate pre-trained model for that fidelity level. This dynamic adaptation allows the system to maintain high accuracy across varying conditions while avoiding the need for a single complex universal model.

The patent introduces an intermediary component (fidelity detection module) that assesses the fidelity of incoming data and mediates the selection of the appropriate ML model. This intermediary layer simplifies the overall system architecture by providing a clear decision-making mechanism for model selection, rather than requiring complex interactions between multiple models.