Inference Model Synthetic Data Generation for Pipeline Reliability

The patent introduces an inference model as an intermediary component between the data pipeline and inaccessible data. When data is unavailable due to user limitations or regulatory constraints, the inference model generates synthetic data that substitutes for the missing information, allowing the pipeline to continue operating without direct access to the original data. This mediator resolves the contradiction by enabling data availability while respecting privacy and compliance constraints.

The patent creates copies of inaccessible data through synthetic data generation. Instead of requiring direct access to the original data that is restricted by user preferences or regulations, the system generates synthetic copies that replicate the necessary data characteristics and structures. These copies allow downstream pipeline operations to proceed while the original restricted data remains inaccessible, thus maintaining both availability and compliance.

The patent trains a single inference model to handle multiple types of user-selected limitations and regulatory constraints. Rather than creating separate specialized models for each limitation type, the universal model is trained to recognize and generate appropriate synthetic data replacements for various restriction scenarios. This multi-functional approach maintains high supplementation accuracy while significantly reducing the number of models required and the associated computational complexity.

The patent merges the functionality of multiple specialized inference models into a single unified model. By combining the learning objectives and training data from various limitation types into one comprehensive model, the system achieves the data supplementation capabilities of multiple models while using only one computational resource. This merging strategy reduces device complexity while preserving the accuracy needed for different user limitation scenarios.