2D Outcome Dataset for Ensemble ML False Positive Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches using ensembles of machine learning models struggle to accurately identify specific input sub-datasets likely depicting a finding, often resulting in a high number of false positives due to noisy outcomes from individual models.
Innovation Solution
The method involves creating a two-dimensional outcome dataset from the outputs of an ensemble of machine learning models, applying a filter function such as a convolutional neural network (CNN) or image processing filters, and analyzing the filtered dataset to pinpoint specific input sub-datasets likely depicting the finding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ensembles of machine learning models are used to analyze datasets, then the accuracy of detecting findings is improved, but the number of false positives increases
Solution Approach 1:
The patent transforms the outcomes of multiple ML models into a two-dimensional outcome dataset where one dimension represents different ML models and the other dimension represents different input sub-datasets. This dimensional transformation enables the application of image processing filters (such as convolutional neural networks) to the outcome data, allowing spatial filtering techniques to reduce false positives while maintaining detection accuracy. The 2D representation converts the problem from a simple classification task into a spatial pattern recognition task.
Solution Approach 2:
The patent introduces an intermediary processing stage between the ML model outcomes and the final finding detection. A filter function (such as a CNN or image processing filter) is applied to the 2D outcome dataset to smooth noisy predictions and reduce false positives. This intermediary filter acts as a mediator that processes the raw model outputs before final interpretation, thereby improving reliability without sacrificing the detection accuracy provided by the ensemble approach.
2Measurement precision
If multiple ML models are applied to each input sub-dataset, then the ability to detect findings is improved, but the computational complexity increases
Solution Approach 1:
By organizing the outcomes of multiple ML models into a two-dimensional dataset structure, the patent enables the use of efficient image processing algorithms and convolutional neural networks. These algorithms are computationally optimized for 2D data and can process the ensemble outcomes more efficiently than traditional approaches would require for arbitrary multi-model outputs, thereby managing computational complexity while maintaining improved detection ability.
Solution Approach 2:
The patent employs a universal filter function (such as a CNN or image processing filter) that can be applied to the 2D outcome dataset regardless of the specific number or type of ML models used in the ensemble. This universal approach allows the same filtering mechanism to work across different ensemble configurations, simplifying the overall system architecture and reducing computational complexity compared to model-specific processing approaches.
Data Source
AI summary
There is provided a computer-implemented method of identifying a finding in an input sub-dataset of an input dataset using an ensemble of machine learning (ML) models, comprising: obtaining outcomes of an ensemble of ML models generated in response to feeding each input sub-dataset of the input dataset into each ML model of the ensemble trained to generate an outcome indicative of likelihood of a finding depicted in a respective input sub-dataset, creating a two dimensional (2D) outcome dataset storing outcomes of the ensemble, wherein a first dimension denotes the respective ML model and a second dimension denotes the respective input sub-dataset, computing a filtered 2D dataset by applying a filter function to the 2D outcome dataset, and analyzing the filtered 2D dataset to identify specific input sub-dataset(s) likely depicting the finding.