Distributed Neural Network for Predictive Path Modeling

The patent segments the monolithic analytical computation into distributed microservices that can be executed independently and in parallel across multiple servers. Each microservice handles specific prediction tasks, allowing the overall computation to be divided into smaller, manageable units that can be processed concurrently, thereby reducing total computation time while maintaining prediction accuracy.

The patent replaces traditional mechanical analytical computation with a neural network-based system. The neural network is trained once on historical data and then performs predictions through efficient forward propagation, substituting the computationally intensive repeated analytical calculations with a pre-trained model that provides both speed and accuracy.

The patent introduces a server infrastructure as an intermediary between the neural network model and user equipment. The server hosts the trained model and handles computationally intensive operations, while user equipment only needs to send requests and receive predictions. This intermediary architecture allows standard user equipment to access advanced prediction capabilities without requiring oversized local hardware.

The patent creates a digital copy of the complex prediction model on remote servers, allowing multiple users to access the same model instance simultaneously. Instead of each user needing their own oversized equipment, the model is copied and deployed on server infrastructure that can serve multiple clients, reducing individual device requirements while maintaining prediction capability.

The patent performs preliminary training of the neural network model on complete historical datasets, including handling of missing values during the training phase. Once trained, the model can perform incremental updates by processing new data through efficient forward propagation without requiring complete recalculation. The training phase handles data completeness requirements, while the prediction phase can operate with partial data.

The patent changes the operational parameters of the prediction system by using a neural network that can accept partial inputs and provide probabilistic outputs. The model's parameters are optimized during training to handle missing values, and during operation, it can process incremental updates with changing data conditions without requiring complete recalculation, thereby improving update efficiency while maintaining reliability through the model's robustness.