Balanced Synthetic Data Generation via GAN Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Imbalanced synthetic training data can lead to biased machine learning models in communication systems, particularly in optimizing resource utilization and detecting intrusions, as existing methods fail to generate sufficiently balanced datasets for effective training.

Innovation Solution

An apparatus and method utilizing competing generative adversarial networks to iteratively refine synthetic data generation by adjusting trainable parameters based on feedback from a discriminator model, ensuring accuracy meets a preset threshold and an end criterion, thereby producing balanced training data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If synthetic training data is generated to obtain enough training data, then the quantity of training data is improved, but the balance of training data deteriorates

Engineering Contradiction:
Improvequantity of training dataVSAvoidbalance of training data
Core Design Contradiction:
Quantity of substanceVSStability of the object's composition

Solution Approach 1:

The patent implements a feedback mechanism where a discriminator model evaluates the quality and distribution of synthetic data generated by a generator model. The discriminator's feedback on data distribution balance guides the generator to produce more balanced synthetic data, resolving the contradiction between quantity and balance of training data.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The discriminator model serves as an intermediary between the generator and the training process. It mediates the data generation process by evaluating whether synthetic data maintains proper distribution balance, allowing the system to generate sufficient quantity of data while preserving balance through the discriminator's guidance.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of manufacture

If imbalanced training data is used, then the training process is simplified, but the model accuracy deteriorates

Engineering Contradiction:
Improvetraining process simplicityVSAvoidmodel accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The system performs self-service by automatically generating balanced synthetic data through the generator-discriminator framework. The discriminator evaluates data balance and provides feedback, enabling the system to self-correct imbalances without manual intervention, thus maintaining both simplicity and accuracy.

Inventive Principle:
Principle #25Self-service

3Productivity

If synthetic data is generated without balancing, then the productivity is improved, but the reliability of trained models deteriorates

Engineering Contradiction:
Improvedata generation speedVSAvoidmodel reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system performs preliminary action by generating and evaluating synthetic data through the discriminator before final training. This preliminary evaluation ensures data balance is established beforehand, allowing fast generation while maintaining reliability through pre-validated data quality.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4407522A1Providing balanced training data
Publication Date: 2024.07.31 NOKIA SOLUTIONS & NETWORKS OY
  • EP4407522A1 patent drawingFigure 1~2
  • EP4407522A1 patent drawingFigure 3~6
  • EP4407522A1 patent drawing

AI summary

Two competing machine learning based models, a first model for generating synthetic data, and a second model for classifying input data to synthetic data and real data, are trained by training the second model until its accuracy meets a preset rule, and then the first model is trained. After training the first model, training of models is repeated until an end criterium is met.