AI Training Data Tags and Codes for Tracing Incorrect Answers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI training methods struggle to accurately identify and correct incorrect training data, which affects model accuracy, reliability, and limits its application potential in various industries due to the difficulty in tracing the source of incorrect answers, especially in large-scale data sets.
Innovation Solution
An AI training method that adds tag data and coded data to training data to ensure the training process remains unaffected, allowing for the identification and correction of incorrect training data by querying coded data corresponding to incorrect answers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional AI training methods are used without additional tag data and coded data, then the training process is simple, but it becomes difficult to identify and correct incorrect training data
Solution Approach 1:
The training data is segmented into multiple components: original training data, tag data (identifying data source, time, etc.), and coded data (hash codes for quick identification). This segmentation allows the system to maintain simple training processes while enabling precise identification of incorrect data through the separate identification components.
Solution Approach 2:
Tag data and coded data act as intermediaries between the training data and the identification/correction process. These intermediary elements carry identification information without interfering with the actual training content, enabling the system to trace and correct incorrect data without complicating the training workflow.
2Productivity
If tag data and coded data are added to training data for identification, then incorrect training data can be identified efficiently, but the data structure becomes more complex
Solution Approach 1:
Tag data and coded data are prepared in advance during the data preparation phase, before the actual training process begins. This preliminary action embeds identification information into the training data structure upfront, enabling rapid identification of incorrect data during testing without requiring complex real-time analysis during training or evaluation.
Solution Approach 2:
Instead of modifying the original training data structure fundamentally, the system creates copies or additions of identification data (tag data and coded data) that parallel the original data. These copied identification elements can be processed separately to identify incorrect data without requiring complex integration with the main training workflow.
3Loss of information
If the AI model is trained with additional identification data, then the source of incorrect answers can be traced, but the training process may be affected
Solution Approach 1:
The identification function is extracted from the main training process. Tag data and coded data are separated from the core training content, allowing the model to learn from the training data while the identification system independently tracks data sources. This extraction ensures that identification requirements do not interfere with or destabilize the training process.
Solution Approach 2:
The system adds identification information in a separate dimension (metadata layer) rather than integrating it into the core training dimensions. The tag data and coded data exist in an additional dimensional space that parallels the training data, enabling traceability without affecting the stability of the original training process.
Data Source
AI summary
The disclosure describes an artificial intelligence training method for identifying incorrect training data and an artificial intelligence correcting method using the same. When training an artificial intelligence (AI) model, a plurality of training data along with their corresponding tag data and coded data are input, thereby ensuring that the training process of the AI model remains unaffected. When the AI model is used to obtain an incorrect answer, the tag data are obtained based on the coded data. A piece of training data referenced by the incorrect answer is identified using the tag data, thereby efficiently updating the original incorrect training data.