Open Set Recognition via Pre-stored Classification Criteria
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing open set recognition technologies face challenges in accurately detecting data of unknown classes in target data sets when the training data set is not available, as re-training the classification model is hindered by the inability to determine which data items belong to unknown classes.
Innovation Solution
A non-transitory recording medium stores a program that identifies data items with a significant degree of contribution to changes in the classification criterion by re-training the classification model using a different data set, detecting items where loss reduction occurs, and flagging them as unknown classes based on specific properties related to movement distances and loss changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If re-training is performed to improve unknown class detection accuracy, then detection accuracy improves, but the process becomes infeasible when training data is unavailable
Solution Approach 1:
The patent performs preliminary actions by storing multiple classification criteria (from different training data sets) in advance. When unknown class detection is needed, the system compares the input data against these pre-stored criteria without requiring re-training, thus avoiding the need for available training data while maintaining detection capability.
Solution Approach 2:
The patent creates copies of classification criteria from different training data sets and stores them for later use. Instead of re-training the original model, the system uses these pre-created copies (classification criteria) to perform unknown class detection, eliminating the dependency on original training data.
2Measurement precision
If classification criteria are changed through re-training, then unknown classes can be detected, but the complexity of determining which data belongs to unknown classes increases
Solution Approach 1:
The patent introduces classification criteria from different training data sets as intermediaries. Instead of directly determining which data belongs to unknown classes (which is complex), the system uses these intermediate classification criteria to indirectly identify unknown classes by finding data that doesn't fit any of the pre-stored criteria.
Solution Approach 2:
The system performs preliminary classification using multiple pre-stored criteria before attempting to identify unknown classes. This preliminary action simplifies the overall process by first filtering data through known criteria, making the subsequent unknown class detection more straightforward.
3Adaptability or versatility
If the classification model is re-trained with target data, then the model adapts to new data, but the original classification capability is lost when training data is unavailable
Solution Approach 1:
The patent applies local quality by maintaining different classification criteria for different data sets. Instead of uniformly re-training the entire model (which would lose original capabilities), the system locally adapts by using specific pre-stored criteria that correspond to different training data sets, preserving original classification capabilities while adapting to new data when needed.
Solution Approach 2:
The patent changes parameters by switching between different pre-stored classification criteria instead of fundamentally re-training the model. This allows the system to adapt to different data sets by selecting appropriate pre-stored criteria, maintaining reliability of original capabilities while achieving adaptability to new data scenarios.
Data Source
AI summary
An information processing method comprising: for a classification model for classifying input data into one or another of plural classes that was trained using a first data set, identifying, in a second data set that is different from the first data set one or more items of data having a specific datum of which a degree of contribution to a change in a classification criterion is greater than a predetermined threshold, the classification criterion being a classification criterion of the classification model during re-training based on the second data set; and, from among the one or more items of data, detecting an item of data, for which a loss reduces for the classification model by change to the classification criterion by re-training based on the second data set, as an item of data of an unknown class not contained in the plural classes.