Normalized Classification Scores for Digital Object Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital classification systems face inaccuracies in classifying digital objects due to variations in the amount of training data, leading to skewed classification scores and incorrect identifications, especially when comparing multiple classification scores for objects with differing numbers of tagged training images.
Innovation Solution
The system normalizes classification scores based on the number of tagged digital training images, using probability functions to transform scores into relative probabilities, thereby accurately classifying unknown digital objects across multiple potential classifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of tagged digital training images is increased to improve classification accuracy, then more training data is available, but classification scores become skewed and lead to incorrect identifications
Solution Approach 1:
The patent transforms classification scores by applying normalization functions that adjust the scores based on the number of training images. This parameter transformation converts raw scores into normalized values that account for training data variations, resolving the skew caused by different training set sizes while maintaining the benefit of having multiple training images.
Solution Approach 2:
The patent introduces a normalization function as an intermediary between the raw classification score and the final classification decision. This intermediary component processes the score to eliminate the bias introduced by varying training data amounts, allowing accurate comparison across different classification models without requiring equal training set sizes.
2Adaptability or versatility
If multiple classification scores are generated for different known objects, then more classification options are provided, but comparing these scores becomes inaccurate due to variations in training data amounts
Solution Approach 1:
The patent applies a normalization transformation to each classification score based on the number of training images used for that specific classification model. This parameter change enables accurate comparison of scores across different objects even when their training data amounts differ, preserving the versatility of multiple classification options while ensuring reliable score comparison.
Solution Approach 2:
The normalization function creates an equipotential basis for comparing classification scores by adjusting all scores to a common reference frame that accounts for training data variations. This allows fair comparison across different classification models regardless of their individual training set sizes, enabling accurate identification among multiple candidates.
3Adaptability or versatility
If training data amount varies across different known objects, then diverse object coverage is achieved, but classification scores cannot be accurately compared
Solution Approach 1:
The patent transforms the classification score parameter by applying a normalization function that incorporates the training image count. This transformation maintains the ability to classify diverse objects with varying training data amounts while correcting the reliability issue by adjusting scores to a standardized scale that enables accurate comparison across all objects.
Data Source
AI summary
The present disclosure is directed toward systems and methods that enable more accurate digital object classification. In particular, disclosed systems and methods address inaccuracies in digital object classification introduced by variations in classification scores. Specifically, in one or more embodiments, disclosed systems and methods generate probability functions utilizing digital test objects and transform classifications scores into normalized classification scores utilizing probability functions. Disclosed systems and methods utilize normalized classification scores to more accurately classify and identify digital objects in a variety of applications.