Boundary-Structured Rulesets for Bounded-Error Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional mapping techniques for multiple input values to output values, such as in classification and decision making, are resource-intensive and prone to overfitting, underfitting, and non-linear decision boundaries, leading to misclassification errors and inefficiencies, especially when dealing with complex systems like engineered systems, and require complete regeneration of mappings with additional data.
Innovation Solution
The use of boundary-structured ruleset (BSR) to refine mappings by converting input records into labelled subregions, merging them based on approximation bounds, and generating a ruleset that provides approximate mappings with bounded errors, using local statistical smoothness and per-attribute approximation bounds to ensure efficient and accurate mapping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional mapping techniques are used for classification and decision making, then mapping accuracy can be achieved, but computational resources and processing time are excessively consumed
Solution Approach 1:
The patent segments the input space into discrete hyperrectangular regions defined by boundary values. Each region is characterized by a centroid point, transforming continuous mapping into discrete region-based mapping. This segmentation reduces computational complexity by avoiding exhaustive processing of continuous input spaces while maintaining mapping accuracy through region-based approximation.
Solution Approach 2:
The patent creates simplified representations (centroids) of complex input regions. Instead of processing entire hyperrectangular regions with multiple input values, the system uses centroid points as representative copies that capture essential region characteristics. This copying approach reduces computational resource requirements while preserving the functional mapping relationship.
2Reliability
If traditional mapping techniques are used, then comprehensive data coverage can be achieved, but storage requirements and processing overhead increase significantly
Solution Approach 1:
The patent merges overlapping or adjacent hyperrectangular regions into unified regions when their centroids and boundary definitions allow. This merging reduces the total number of discrete regions that need to be stored and processed, thereby decreasing storage requirements while maintaining comprehensive data coverage through the merged region representations.
Solution Approach 2:
The patent extracts essential characteristics from complex input data by identifying boundary values and computing centroids. Instead of storing complete input-output mappings for all possible input combinations, the system extracts and stores only the essential region-defining parameters (boundaries and centroids), significantly reducing storage requirements while preserving mapping reliability.
3Measurement precision
If traditional mapping techniques are used, then precise decision boundaries can be formed, but the system becomes prone to overfitting and underfitting
Solution Approach 1:
The patent uses approximation bounds that allow partial precision rather than exact decision boundaries. By defining regions with tolerance margins and using centroids as approximate representatives, the system accepts partial precision in exchange for improved robustness. This prevents overfitting to exact boundary conditions while maintaining sufficient decision-making accuracy for practical applications.
4Measurement precision
If traditional mapping techniques are used, then complete mapping regeneration is required with additional data, but this increases design time and runtime
Solution Approach 1:
The patent performs preliminary segmentation of the input space into hyperrectangular regions and computes centroid representations in advance. This preliminary action creates a structured framework that can accommodate additional data points without requiring complete regeneration. New data can be integrated by identifying which pre-defined regions they fall into, significantly reducing both design time and runtime compared to traditional complete regeneration approaches.
Data Source
AI summary
A labelled point within a multidimensional space is input from a set of input records. It is determined whether the labelled point is subsumed by a member hyperregion of the set of hyperregions specified by an BSRS, wherein the BSRS includes a boundary-structured rule corresponding to a hyperregion in the multidimensional space. In the event the labelled point is not subsumed, the BSRS is expanded or the labelled point is eliminated.