Data Classification System for Zero-Shot Learning Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data classification systems face challenges in accurately classifying data into classes that are not present in the training data, known as unknown classes, which affects the accuracy of zero-shot learning techniques.
Innovation Solution
A data classification system that calculates known class likelihoods, selects candidate classes by excluding known classes, and estimates the class based on all-class likelihoods, improving the accuracy of class classification by narrowing down candidate classes and utilizing inter-class similarity calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing zero-shot type data classification techniques are used to classify test data into unknown classes, then class classification can be performed for classes not in training data, but the accuracy of class classification deteriorates due to erroneous estimations
Solution Approach 1:
The patent segments the classification process into two distinct stages: first calculating known class likelihoods for classes present in training data, then using these results to select candidate unknown classes. This segmentation allows the system to leverage reliable known class information while systematically exploring unknown class possibilities, thereby improving overall classification accuracy without sacrificing adaptability to new classes.
Solution Approach 2:
The patent performs preliminary calculation of known class likelihoods before proceeding to unknown class classification. By first determining the likelihood of known classes and using this information to filter and select candidate unknown classes, the system establishes a solid foundation that reduces erroneous estimations in subsequent unknown class classification, thus improving measurement precision while maintaining versatility.
2Adaptability or versatility
If all classes are considered for classification without filtering, then comprehensive classification coverage is achieved, but processing efficiency deteriorates due to increased computational complexity
Solution Approach 1:
The patent extracts and utilizes known class likelihood information from training data to identify and select only the most relevant candidate unknown classes. By taking out the essential information from known classes and applying it to filter the unknown class candidates, the system maintains comprehensive classification coverage while significantly reducing the number of classes that require intensive computational processing, thereby improving processing efficiency.
Solution Approach 2:
The patent changes the parameter of class selection by using known class likelihood values as a filtering criterion. Instead of processing all unknown classes uniformly, the system adjusts the selection parameter to focus computational resources on candidate classes with higher likelihoods, thus maintaining comprehensive coverage of potential unknown classes while improving processing efficiency through parameter-based filtering.
Data Source
AI summary
A data classification system calculates, for each known class, which is a class indicated in training data used for performing learning of class classification, a known class likelihood indicating a likelihood of target data belonging to the known class among all known classes. The data classification system selects, as candidates for a class to which the target data belongs, classes in which at least one of the known classes is excluded based on the known class likelihood among all classes in the class classification. The data classification system calculates, at least for each of the classes included in the candidates, an all-class likelihood indicating a likelihood of the target data belonging to the class among all the classes. The data classification system estimates the class to which the target data belongs as any one of the classes among the candidates, based on the all-class likelihood.