Continuously Generalized Ordinal Regression Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ordinal regression methods are too restrictive and inflexible, particularly when features fall outside expected modeling assumptions, leading to poor performance on various data sets.
Innovation Solution
The introduction of continuously generalized ordinal logit (coGOL) with adjustable model flexibility and a regularization step to control bias-variance tradeoff, allowing for non-parallel hyperplane slopes in ordinal classification bins, and extension to deep learning using the all-thresholds loss function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If threshold-based models with parallel hyperplanes are used, then the model structure is simple, but the model flexibility and accuracy deteriorate when features fall outside expected modeling assumptions
Solution Approach 1:
The patent segments the single linear mapping into multiple independent linear mappings, one for each class. This allows each class to have its own hyperplane with potentially different slopes, enabling the model to capture non-parallel relationships in the data while maintaining a structured approach through separate class-specific mappings.
Solution Approach 2:
The patent introduces dynamic flexibility by allowing the hyperplane slopes to vary across different classes rather than being fixed and parallel. This dynamic structure adapts to the underlying data relationships, enabling the model to adjust its decision boundaries based on class-specific patterns in the ordinal regression problem.
2Adaptability or versatility
If generalized coefficients with separate linear mappings for each class are introduced, then model flexibility improves, but model complexity increases
Solution Approach 1:
The patent applies a universal loss function framework that works across different ordinal regression scenarios. The continuously generalized ordinal logit (coGOL) loss function serves as a multi-functional objective that can handle various data configurations and ordinal relationships, reducing the need for problem-specific model variations while maintaining flexibility.
Solution Approach 2:
The patent changes the parameters of the loss function to continuously generalize the ordinal logit approach. By introducing a continuously adjustable parameter that controls the generalization degree, the model can adapt its complexity level based on the specific problem requirements, balancing flexibility and computational efficiency through parameter optimization rather than structural complexity.
3Device complexity
If cumulative link loss models are used, then the model structure is constrained, but training efficiency deteriorates on data sets with features outside expected assumptions
Solution Approach 1:
The patent replaces the static cumulative link loss structure with a dynamic continuously generalized ordinal logit loss function. This dynamic loss function adapts to the data characteristics during training, allowing the model to efficiently learn from diverse data sets with features outside traditional assumptions by continuously adjusting its optimization landscape rather than being constrained by fixed cumulative link structures.
Data Source
AI summary
A method and system for configuring a computer for data classification using ordinal regression includes: receiving and storing a data set having data with a plurality of data features that have an ordinal relationship; generating a plurality of ordinal classification bins based on the relationship of the data features, at least one ordinal classification bin having walls defined by at least two hyperplanes; generating an ordinal regression model of the data set illustrating the data of the data set arranged into the plurality of ordinal classification bins; and tuning the slopes of the walls of the at least one ordinal classification bin based on the relationships between the plurality of data features of the data arranged within the at least one ordinal classification bin such that the slopes of the two hyperplanes defining the walls of the at least one ordinal classification bin are not parallel.