Classifier Generation Using Source Task Data Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for classifying input patterns into specific categories face challenges when data amounts for non-specific categories are insufficient, leading to inaccurate classifications, as they rely on transformation rules dependent on specific data, which can result in improper learning data.
Innovation Solution
An information processing apparatus that obtains learning data from both a target task and a source task, adjusts transformation parameters based on distribution differences, and transforms source task learning data to generate a classifier with improved precision by adding abnormal data from the source task to the target task, even when the target task's abnormal data is scarce.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If learning data from another task is transformed and added to supplement insufficient learning data in the target task, then the amount of learning data is increased and learning can be performed, but the transformation rule becomes dependent on specific data used for obtaining the rule, resulting in insufficient transformation rules and possible improper learning data
Solution Approach 1:
The patent performs preliminary actions by obtaining multiple learning data sets from the source task before transforming them. Specifically, it obtains first learning data with first category labels and second learning data with second category labels, then uses these pre-obtained datasets to derive transformation rules that can properly map the source task data to the target task without relying on specific dependent data.
Solution Approach 2:
The patent segments the learning data into different categories (first category and second category) and processes them separately. By dividing the learning data into these segments and obtaining transformation rules for each segment independently, the system avoids the problem of transformation rules being dependent on specific data while maintaining the ability to transform all source task data properly.
2Productivity
If a classifier is generated using insufficient non-specific category patterns, then the classifier can be generated, but it is liable to erroneously determine that non-specific category patterns not used for generation are specific categories
Solution Approach 1:
The patent merges the transformed source task learning data with the target task learning data to create a combined learning dataset. This combined dataset is then used to generate the classifier, ensuring that both specific and non-specific category patterns are adequately represented in the training data, which prevents the classifier from erroneously classifying unseen non-specific patterns as specific categories while maintaining efficient generation.
Data Source
AI summary
On the basis of a difference between first distribution regarding target task learning data as a plurality of learning data which belongs to a first category of a first task as a target task and second distribution regarding a plurality of learning data which belongs to the first category of source task learning data as learning data which belongs to the first category of a second task different from the first task, a transformation parameter for transforming the source task learning data is adjusted, the source task learning data is transformed based on the adjusted transformation parameter, and a classifier regarding the first task is generated based on the transformed source task learning data and the target task learning data.