Learning Device for Voice Recognition Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing models in contact center voice recognition systems require large amounts of training data and suffer from accuracy deterioration when new data is added, especially when attribute information is not considered, leading to increased storage costs and potential data obsolescence due to privacy restrictions.
Innovation Solution
A learning device and method that processes new training data based on attribute information to create a new model by dividing the data into subsets and incrementally learning each subset, thereby maintaining estimation accuracy and updating model parameters gently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If all existing training data and new training data are used for model learning, then the model can be updated with new information, but it takes time to learn the model and evaluate accuracy, and storage costs increase
Solution Approach 1:
The patent segments the training data into existing training data and new training data, and segments the learning process into fine-tuning (using only new training data) and full relearning (using all training data). This allows selective updating based on resource constraints, reducing time and storage costs while maintaining model adaptability.
2Productivity
If fine tuning is performed by additional learning of new training data for an existing model, then the model can be updated efficiently, but the tendency of the learned existing training data is forgotten and estimation accuracy deteriorates
Solution Approach 1:
The patent introduces a dynamic selection mechanism that determines whether to perform fine-tuning or full relearning based on the degree of difference between existing and new training data. When the difference is small, efficient fine-tuning is used; when the difference is large, full relearning is performed to prevent accuracy deterioration, thus adaptively balancing efficiency and precision.
3Measurement precision
If existing training data is kept for model learning, then the model maintains high estimation accuracy for existing data, but data storage cost increases and data may become unavailable due to privacy restrictions
Solution Approach 1:
The patent creates a virtual representation of existing training data through the existing model parameters and knowledge embedded in the model. Instead of physically storing and reusing the original training data, the model captures the essential patterns and tendencies, allowing accurate predictions without retaining the actual training data, thus reducing storage costs and privacy concerns.
4Measurement precision
If a large amount of training data is used to set estimation accuracy to a level for practical use, then high estimation accuracy can be obtained, but the complexity of data management and processing increases
Solution Approach 1:
The patent segments the training data management into separate handling of existing training data and new training data. The system evaluates the degree of difference between these segments and selectively processes only the necessary portion (either just new data for fine-tuning or all data for full relearning), significantly reducing data management complexity while maintaining high estimation accuracy.
Data Source
AI summary
A learning device (10) according to the present disclosure includes a data set division unit (11) as a training data processing unit and a divided data set learning unit (12) as a model learning unit. The data set division unit (11) divides a new training data set into a plurality of divided data sets on the basis of attribute information. After performing model learning processing using an existing model as a learning target model, the divided data set learning unit (12) creates a new model by repeating the model learning processing until all the divided data sets are learned using a learned model created by the model learning processing as a new learning target model.