Query-Oriented Event Recognition Using Reusable Inference Engines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI inference modules are time-consuming and costly to train, and cannot be reused across different application scenarios, leading to increased development costs and reduced efficiency.
Innovation Solution
A query-oriented event recognition system with a capturing unit, analysis and inference engine assembly, synchronization processing unit, and event definition unit, which utilizes multiple inference engines to analyze and reason data, integrate information, and quickly identify events through query statements, reducing the need for re-training and re-analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an AI inference module is trained to a sufficient level of understanding, then event identification accuracy is improved, but development time and cost increase significantly
Solution Approach 1:
The system performs preliminary actions by capturing and storing raw data flows with time points in advance. When event identification is needed, the pre-captured data can be quickly queried and analyzed without requiring time-consuming re-collection or re-processing, thus reducing development and analysis time while maintaining accuracy
Solution Approach 2:
The system segments the event identification process into distinct modules: data capture, inference engine analysis, synchronization, and event definition querying. This segmentation allows each module to be optimized independently and enables parallel processing, reducing overall development time while maintaining high identification accuracy
2Measurement precision
If an AI inference module is trained for a specific application scenario, then event identification accuracy for that scenario is improved, but the module cannot be reused for new scenarios, increasing development cost
Solution Approach 1:
The system implements universality by designing a scenario-independent data capture and inference engine architecture. The capturing unit stores raw data flows with time points that can represent multiple different scenarios. The inference engines and event definition unit can be configured through query statements to identify events across different application scenarios without requiring retraining, enabling one system to serve multiple functions and scenarios
Solution Approach 2:
Instead of creating new trained models for each scenario, the system captures and stores data representations that can be copied and reused across scenarios. The inference results and synchronized data can be queried and reused for different event identification needs, eliminating the need to reprocess original data and reducing development costs for new scenarios
3Measurement precision
If a new inference module is trained for each new application scenario, then event identification accuracy is improved, but development cost and time increase significantly
Solution Approach 1:
The system implements universality by designing a scenario-independent data capture and inference engine architecture. The capturing unit stores raw data flows with time points that can represent multiple different scenarios. The inference engines and event definition unit can be configured through query statements to identify events across different application scenarios without requiring retraining, enabling one system to serve multiple functions and scenarios
Solution Approach 2:
The system changes parameters by using query statements with different criteria to identify different events from the same inferred data set. Instead of changing the underlying model or retraining, the system adjusts the query parameters (event definitions, time point ranges, data patterns) to adapt to different scenarios, maintaining accuracy while reducing development cost
Data Source
AI summary
A query-oriented event recognition system includes a capturing unit, an analyzing inference engine assembly, a synchronization processing unit and an event definition unit. The capturing unit is configured to capture at least one data flow including a plurality of time points and data, and each time points is corresponding to a data. The analysis and inference engine assembly includes a plurality of inference engines. The analysis and inference engine assembly is configured to analyze and reason the data of the time points by the inference engines to generate at least one inference data corresponding to the data at each time point. The synchronization processing unit is configured to generate an inference data set corresponding to each time point according to the data flow and the inference data. The event definition unit is configured to find an event query result matching up with the inference data set through a query statement.