Distributed DNN Ensemble Prediction for XR User Inputs at the Edge

The patent divides the prediction system into multiple distributed deep neural networks operating independently on the edge cloud, each handling specific aspects of user input prediction. This segmentation allows the system to achieve higher accuracy through ensemble predictions while managing complexity by distributing computational loads across separate network instances.

The patent introduces an intermediary layer that coordinates between multiple distributed DNNs and the XR device, managing data flow, prediction aggregation, and model updates. This intermediary structure simplifies the overall system architecture by providing a standardized interface while enabling complex distributed processing.

The patent implements preliminary action by pre-training distributed neural networks offline with extensive environmental and user interaction data before deployment. This pre-training enables the models to adapt quickly to real-time conditions with minimal additional computational energy, as the heavy lifting of learning fundamental patterns has already been completed.

The system employs periodic action by updating and retraining the distributed DNNs at scheduled intervals rather than continuously. This approach maintains prediction reliability by periodically adapting to new environmental conditions and user behaviors while significantly reducing computational energy consumption compared to continuous real-time retraining.