AI Model Training Region of Interest Definition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI model training methods are inefficient due to reliance on user-specified or self-determining regions of interest, leading to either inefficiency or performance degradation, as they often fail to capture sufficient context, resulting in slow training speeds and poor model performance.
Innovation Solution
A processor-implemented method that automatically defines and updates regions of interest by summarizing context information around these areas, allowing for high-speed training without performance degradation by iteratively refining the region boundaries based on prediction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If user-specified regions of interest are used, then all labels are correct and independent of external information, but training efficiency becomes too slow to be practical for large documents
Solution Approach 1:
The system automatically determines regions of interest without requiring user specification. The automated region determination algorithm analyzes document structure and label positions to identify optimal regions, enabling the system to serve itself rather than requiring manual user input for each region boundary.
Solution Approach 2:
The patent dynamically adjusts region boundary parameters based on document characteristics and label positions. By changing the fixed user-specified boundary parameters into adaptive parameters that respond to document content, the system achieves both accuracy and efficiency.
2Productivity
If self-determining regions of interest are used, then training speed increases, but critical information is lost causing severe degradation in AI model performance
Solution Approach 1:
The system incorporates feedback mechanisms where the automated region determination is continuously refined based on training results and performance metrics. This feedback loop allows the system to adjust region boundaries to maintain optimal performance while preserving training speed advantages.
Solution Approach 2:
The region boundaries are made dynamic rather than static, allowing them to adapt during training based on the specific document content and label distribution. This dynamic adjustment enables the system to optimize both speed and performance for each specific training instance.
3Reliability
If region of interest boundaries are expanded to include more context, then AI model performance improves, but training time increases significantly
Solution Approach 1:
The patent segments the document into multiple regions of interest based on label positions and document structure. Instead of processing the entire document as one unit, the system divides it into manageable segments that can be processed in parallel or sequentially, reducing overall training time while maintaining performance.
Solution Approach 2:
The system uses partial action by focusing training only on the necessary regions surrounding labels rather than the entire document. This selective approach processes only the essential information needed for accurate label prediction, avoiding the time cost of processing irrelevant portions of the document.
Data Source
AI summary
A processor-implemented method includes (i) defining a region of interest ranging between a first and second boundary location for each label in the M documents that comprise N labels, (ii) summarizing information, in a selected document, from a first content location to the first boundary location of the region of interest to obtain a first summary that represents context information from the first content location to the first boundary location of the region of interest, (iii) summarizing information, in the selected document, from a second content location to the second boundary location to obtain a second summary that represents context information from the second boundary location to the second content location, (iv) performing training of the AI model including restricting training data from the M documents based on the region of interest, and (v) extracting the target data from the M documents using trained AI model.