Training Apparatus Adjusting Data Scores for Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for updating training data in inference models, such as neural networks, are time-consuming and prone to accuracy decreases due to imbalanced data selection, particularly caused by seasonal fluctuations, requiring manual intervention by skilled personnel.
Innovation Solution
A training apparatus and method that automatically adjusts scores of data in a classification space to distribute training data more evenly, preventing concentration near identification boundaries, thereby allowing for efficient and accurate model updates without significant human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is added one by one by manual selection, then inference accuracy can be improved, but the time required to update model parameters increases significantly
Solution Approach 1:
The system performs preliminary actions by automatically selecting and preparing training data before model updating. The data selection unit pre-processes and ranks data points based on their importance to the model, so that when updating is needed, the data is already organized and ready for efficient processing, reducing the overall time required.
Solution Approach 2:
The system enables self-service by automatically selecting training data without requiring manual intervention from skilled personnel. The data selection unit autonomously evaluates data importance and selects appropriate training data, allowing the system to update model parameters efficiently without human involvement in the data selection process.
2Reliability
If training data is selected manually by skilled personnel, then data quality can be maintained, but the process becomes time-consuming and labor-intensive
Solution Approach 1:
The data selection unit performs self-service by automatically evaluating and selecting training data based on predefined criteria and model requirements. It autonomously determines which data points are most important for training, eliminating the need for manual review while maintaining data quality through algorithmic selection processes.
Solution Approach 2:
The system changes parameters by using automated selection criteria based on data characteristics, model performance metrics, and importance weights. Instead of manual selection, the system adjusts and optimizes selection parameters algorithmically to identify the most valuable training data, improving processing efficiency while maintaining quality.
3Productivity
If training data is added without considering data distribution balance, then the update process becomes faster, but inference accuracy decreases due to data imbalance
Solution Approach 1:
The system changes parameters by adjusting data selection criteria to account for distribution balance. The data selection unit modifies selection weights and importance parameters to ensure that training data represents the true data distribution, preventing imbalance while maintaining fast update speeds through automated processing.
Solution Approach 2:
The system implements feedback by continuously monitoring model performance and data distribution characteristics. The feedback mechanism allows the data selection unit to adjust its selection criteria based on actual model behavior and performance metrics, ensuring that training data maintains balance and improves accuracy while enabling rapid updates.
Data Source
AI summary
According to an embodiment, a training apparatus is a training apparatus training an inference model performing predetermined inference based on data, the training apparatus changes, according to a predefined distance for a selected piece of data selected based on predefined scores given to a plurality of pieces of data in a predefined classification space, the predefined score of a piece of data other than the selected piece of data, gives labels in the predefined classification space to pieces of data including the piece of data the predefined score of which is changed, builds training data from the pieces of data to which the labels are given, and updates the inference model using the training data.