ROC Curve Area Minimization for Stable Two-Class Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The learning method using pAUC tends to converge to locally optimal solutions when updating parameters by hill climbing, resulting in unstable high accuracy for two-class classification.

Innovation Solution

The method sets parameters in the score function to minimize the area over the ROC curve within a given false positive rate threshold, using a hill descending method with a differentiable function that approximates the area, rather than maximizing partial AUC, to avoid locally optimal solutions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the learning method using pAUC is employed to set parameters by maximizing partial AUC, then the false positive rate can be limited to not more than a predetermined threshold, but the parameter solution tends to be a locally optimal solution resulting in unstable classification accuracy

Engineering Contradiction:

Improvefalse positive rate controlVSAvoidclassification accuracy stability

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

Instead of maximizing the partial AUC area as in conventional methods, the patent inverts the objective to minimizing the area over the ROC curve within the same false positive rate threshold. This inversion transforms the optimization landscape, preventing convergence to locally optimal solutions while maintaining reliable false positive rate control and achieving stable high classification accuracy.

Inventive Principle:

Principle #13The other way round (Inversion)

2Object-affected harmful factors

If hill climbing method is used to update parameters by maximizing pAUC, then the false positive rate can be constrained, but the optimization converges to locally optimal solutions

Engineering Contradiction:

Improvefalse positive rateVSAvoidparameter solution stability

Core Design Contradiction:

Object-affected harmful factorsVSReliability

Solution Approach 1:

The patent applies hill descending method to minimize the area over the ROC curve instead of using hill climbing to maximize pAUC. This inverted optimization approach fundamentally changes the convergence behavior, eliminating the tendency to get trapped in local optima while still constraining the false positive rate within acceptable thresholds.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The patent changes the objective function parameter from maximizing partial AUC to minimizing area over ROC curve. This parameter transformation in the optimization target creates a more favorable optimization landscape that avoids local optima, thereby improving the reliability and stability of the parameter solutions obtained through iterative optimization.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240419765A1Learning device, learning method, and learning program and sorting device, sorting method, and sorting program

Publication Date: 2024.12.19 NEC CORP

AI summary

To carry out a learning technique in which a solution of a parameter is not likely to be a locally optimal solution, a learning apparatus includes: a learning means for setting a parameter (θ) included in a score function (f) for carrying out two-class classification of data, the learning means setting the parameter (θ) such that, in a square having a horizontal axis representing a false positive rate and a vertical axis representing a true positive rate, an area of a region in which the false positive rate is not more than a given threshold is minimized in a region over a receiver operating characteristic (ROC) curve obtained from a training data group.