Information Conversion Using Preconfigured Mapping List

The patent segments the information conversion process into two independent parts: a pre-trained neural network model that handles general conversion tasks, and a separate configurable mapping list that stores specific mapping relationships. This segmentation allows the model to maintain its trained accuracy while the mapping list can be flexibly updated without retraining the entire network.

The patent introduces a mapping list as an intermediary component between the neural network model and the conversion output. This mapping list acts as a mediator that can be independently configured and updated, allowing new mapping relationships to be added without affecting the trained model, thus resolving the contradiction between maintaining model accuracy and enabling flexible updates.

The patent performs preliminary action by pre-training the neural network model once to establish a solid foundation for information conversion. After this preliminary training, the system can adapt to new mapping relationships by simply updating the configurable mapping list, avoiding the need for time-consuming retraining while maintaining adaptability.

The patent creates a separate copy of mapping relationships in the form of a configurable mapping list, which is independent from the trained model parameters. This copying approach allows the system to maintain multiple versions of mapping relationships and switch between them without affecting the original trained model, enabling fast adaptation without retraining time.