Machine Learning Model for Out-Of-Distribution Input Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often struggle with accurately classifying input data belonging to outlier classes due to insufficient training samples, leading to misclassification and loss of trust in their outputs, especially when outlier classes resemble inlier classes.
Innovation Solution
The model is configured to determine whether input data belongs to inlier or outlier classes by calculating scores for each class and comparing the sums of inlier and outlier scores, allowing it to abstain from classification when unsure, thus preventing misclassification and improving trustworthiness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the machine learning model is trained with insufficient samples for outlier classes, then the model can cover more potential classes, but the classification accuracy for outlier classes deteriorates
Solution Approach 1:
The patent segments the classification task into two distinct stages: first determining whether input data belongs to inlier or outlier classes, and then performing fine-grained classification only for inlier classes. This segmentation allows the model to handle diverse classes without sacrificing accuracy on outlier detection, as each stage focuses on a specific aspect of the classification problem.
Solution Approach 2:
The patent introduces an intermediary classification layer that acts as a gatekeeper between raw input and final class prediction. This intermediary stage evaluates whether the input belongs to inlier or outlier classes before allowing progression to fine-grained classification, thereby improving overall reliability by preventing misclassification of outliers as inliers.
2Adaptability or versatility
If the machine learning model attempts to classify all classes including outliers, then the model provides comprehensive coverage, but the reliability of predictions deteriorates due to misclassification
Solution Approach 1:
The patent divides the class set into two distinct segments: inlier classes for which the model is adequately trained and outlier classes for which training is insufficient. This segmentation enables the model to provide comprehensive coverage while maintaining reliability by applying different classification strategies to different segments.
Solution Approach 2:
Instead of attempting to classify all classes directly, the patent inverts the approach by first determining what the input is NOT (i.e., identifying outliers) before classifying what it IS (inliers). This inversion allows the model to maintain reliability by explicitly detecting cases where classification should be avoided.
3Ease of operation
If the machine learning model classifies input data without distinction between inlier and outlier classes, then the processing is simpler, but misclassification errors increase
Solution Approach 1:
The patent segments the classification process into two sequential steps: coarse-grained inlier/outlier classification followed by fine-grained inlier classification. While this adds a step to the process, each step remains conceptually simple and computationally efficient, maintaining ease of operation while significantly improving reliability.
Solution Approach 2:
The patent applies partial classification action by performing detailed fine-grained classification only on inlier classes where the model is confident, while using a simpler coarse-grained approach for outlier detection. This partial application of complex classification reduces overall computational burden while improving accuracy where it matters most.
Data Source
AI summary
A method includes determining, by a machine learning model and based on input data, a feature map that represents learned features present in the input data. The method also includes, for each respective inlier class of a plurality of inlier classes, determining, by the machine learning model and based on the feature map, a corresponding inlier score indicative of a probability that the input data belongs to the respective inlier class. The method additionally includes, for each respective outlier class of a plurality of outlier classes, determining, by the machine learning model and based on the feature map, a corresponding outlier score indicative of a probability that the input data belongs to the respective outlier class. The method further includes determining, based on the inlier scores and the outlier scores, whether the input data corresponds to the plurality of inlier classes or to the plurality of outlier classes.