Generative Model Training Excluding Unauthorized Creator Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing generative models face challenges in managing unauthorized training data elements, as it is difficult to determine their contribution to the model's output and computationally inefficient to remove their impact.

Innovation Solution

Train multiple generative models on different datasets that exclude data from specific creation entities, allowing for quick identification and blocking of models that use unauthorized data, and evaluating their contribution to the output.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If multiple generative models are trained on different datasets excluding specific creation entities, then the ability to manage and block unauthorized data usage is improved, but the device complexity and training time increase

Engineering Contradiction:

Improvecompliance with usage permissionsVSAvoidnumber of trained models

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The training dataset is segmented by creation entity, with each generative model trained on a specific subset excluding particular creation entities. This segmentation enables granular control over which models can be blocked based on their training data composition, directly addressing compliance requirements while managing complexity through organized data partitioning.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Multiple generative models are pre-trained on different dataset configurations before deployment, with each model's training data explicitly excluding specific creation entities. This preliminary action establishes compliance-ready models in advance, allowing rapid blocking or unblocking of specific models based on usage permissions without requiring retraining or complex runtime modifications.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If statistical comparison is performed between target data elements to determine contribution of excluded creation entities, then the precision of identifying unauthorized data influence is improved, but the computational time and resources increase

Engineering Contradiction:

Improvedetection of unauthorized data contributionVSAvoidcomputation time for statistical comparison

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The statistical comparison process replaces manual or mechanical analysis of data element contributions with automated computational methods. By using statistical tests to compare target data elements between models trained with and without specific creation entities, the system efficiently detects unauthorized data influence through algorithmic comparison rather than exhaustive manual analysis, achieving high precision with manageable computational overhead.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20260030479A1Systems and methods for training generative machine learning models

Publication Date: 2026.01.29 NEC CORPOATION OF AMERICA

AI summary

There is provided a computer implemented method of training a generative model, comprising: clustering training data elements each associated with an indication of a creation entity of creation entities, into clusters, each cluster including training data elements associated with one creation entity of the creation entities, generating training dataset by accessing training data elements from a sub-set of the clusters, each training dataset excluding at least one cluster of the clusters associated with at least creation entity, and training generative models on the training datasets, wherein each trained generative model of trained generative models is trained on training data elements that exclude at least one creation entity, wherein a target data element generated by a certain trained generative model in response to an input prompt excludes influence of the excluded at least one creation entity.