Automated Training Data Generation for Machine Learning Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The performance of machine learning models in natural language understanding systems varies significantly based on the configuration of training data sets, making it burdensome for administrators to analyze and prepare effective training data sets.

Innovation Solution

An information processing system that includes a training server for training machine learning models, a response server for outputting responses based on user input, and mechanisms for evaluating model performance, updating training data sets, and retraining models to ensure performance meets predetermined conditions, thereby facilitating the preparation of a training data set.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If administrators manually analyze and prepare training data sets to ensure machine learning model performance, then model performance can be optimized, but the administrative burden and time consumption increase significantly

Engineering Contradiction:
Improvemodel performanceVSAvoidadministrative time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system enables self-service through automated performance evaluation and data generation. The evaluation unit automatically assesses model performance on test data, and the data generation unit automatically creates additional training data when performance is insufficient, eliminating the need for manual administrative intervention while ensuring model performance requirements are met.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary actions by pre-generating candidate training data and pre-evaluating model performance before deployment. The data generation unit prepares additional training data in advance based on performance requirements, and the evaluation unit conducts preliminary performance assessment using test data sets, allowing administrators to deploy models that are already optimized.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If comprehensive training data sets are prepared to improve model performance, then natural language understanding accuracy increases, but the complexity of data preparation and management increases

Engineering Contradiction:
Improvenatural language understanding accuracyVSAvoiddata preparation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system implements feedback through the evaluation unit that continuously monitors model performance on test data and provides feedback to the data generation unit. When performance metrics fall below requirements, the system automatically generates additional training data with correct answers, creating a closed-loop feedback mechanism that systematically improves data quality without increasing preparation complexity.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The evaluation unit acts as an intermediary between the training data set and the machine learning model. It mediates the quality assurance process by automatically evaluating model performance and identifying when additional training data is needed, bridging the gap between data preparation and model performance optimization without requiring direct administrative intervention.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If training data sets are frequently updated to maintain model performance, then model accuracy is maintained, but the processing time and computational resources increase

Engineering Contradiction:
Improvemodel accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system employs periodic action through scheduled performance evaluations using test data sets. The evaluation unit periodically assesses model performance at predetermined intervals, and the data generation unit periodically generates additional training data only when performance degradation is detected, rather than continuously updating data sets. This periodic approach maintains model accuracy while reducing unnecessary processing overhead.

Inventive Principle:
Principle #19Periodic action

Data Source

PatentEP4020339A1Information processing system, information processing method and information processing device
Publication Date: 2022.06.29 RAKUTEN GROUP INC
  • EP4020339A1 patent drawingFigure 1
  • EP4020339A1 patent drawingFigure 2
  • EP4020339A1 patent drawingFigure 3

AI summary

An information processing system obtains a training data set including input data and a label, which is ground truth data for the input data, training a machine learning model on the training data set, inputs test data to the machine learning model trained on the training data set, evaluates whether performance of the machine learning model satisfies a predetermined condition based on an output of the machine learning model to which the test data is entered, updates the training data set when the performance of the machine learning model is evaluated not to satisfy the predetermined condition, and retrains the machine learning model on the updated training data set. The information processing system repeats updating, retraining, and evaluating the data set in response to the evaluation.