Edge Forecasting Model Rollback for Low-Latency Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Edge devices face significant challenges with time to data insights (TDI) and round-trip latencies, particularly in chained forecasting procedures, leading to inaccurate decisions that hinder near-real-time accuracy.
Innovation Solution
Implementing a rollback procedure that interrupts and refreshes forecasting models when reliable state data becomes available, using a rollback netcode framework to manage latency and interdependency issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If chained forecasting procedures are used at edge devices, then comprehensive forecasting capability is achieved, but latency and time to data insights increase significantly
Solution Approach 1:
The system performs preliminary actions by generating predicted data before receiving actual data from dependent forecasting models. The rollback permission is determined in advance, and when actual data arrives, the system can quickly rollback to use the actual data instead of predicted data, thus reducing latency while maintaining comprehensive forecasting capability
Solution Approach 2:
The system implements a feedback mechanism where actual data from dependent forecasting models is continuously monitored and compared with predicted data. When actual data becomes available, the system feedbacks this information to interrupt and refresh the forecasting model, ensuring accuracy while managing latency through the rollback permission threshold
2Productivity
If predicted data is used to initiate forecasting when actual data is not yet received, then forecasting operations can proceed without delay, but accuracy deteriorates
Solution Approach 1:
The system dynamically adjusts the data source for forecasting operations. It transitions from using predicted data (when actual data is not available) to using actual data (when it becomes available), based on real-time conditions. The rollback permission threshold dynamically controls this transition, allowing the system to maintain both speed and accuracy adaptively
Solution Approach 2:
The system changes the parameter of data source quality from predicted to actual data based on the arrival of actual data from dependent forecasting models. This parameter change is controlled by the rollback permission mechanism, which ensures that the system switches to more accurate actual data when available, thereby improving forecasting accuracy without sacrificing operational continuity
3Measurement precision
If rollback permission threshold is set low to ensure accuracy, then forecasting accuracy improves, but the ability to interrupt forecasting operations is reduced
Solution Approach 1:
The system creates a copy mechanism by maintaining both predicted data and actual data pathways. The rollback permission threshold acts as a control parameter that determines when to switch from the predicted data copy to the actual data copy, ensuring accuracy while preserving the ability to interrupt and refresh operations when the threshold is met
Data Source
AI summary
Techniques for rolling back a forecasting operation are disclosed. A service determines a rollback permission for a first forecasting model that generates forecasted data. The rollback permission indicates a commitment threshold. The service generates predicted data that predicts data that has not been received from a second forecasting model and uses the predicted data to initiate generation of the forecasted data. After an initial subset of the forecasted data is generated, the service receives up-to-date data from the second forecasting model. The service determines that the commitment threshold of the rollback permission has not been exceeded despite a remaining subset of the forecasted data still awaiting generation. The service temporarily interrupts the generation of the forecasted data. The service causes the first forecasting model to use the up-to-date data to finish generating the remaining subset of the forecasted data.