Automated Machine Learning with Tokenized Multi-Model Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning systems do not operate on the entire available dataset due to data storage structures, leading to incomplete learning and reliance on random initialization, which makes decisions non-transparent and untraceable.

Innovation Solution

A multi-model database system that uses a document-oriented data model to store raw data without preconceptions, generates tokens from field values, and classifies them as established facts or tentative assumptions, enabling continuous learning and deterministic decision-making.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If data is stored in a database for machine learning training, then data can be reused multiple times, but the database structure requires rigid schemas that limit flexibility

Engineering Contradiction:
Improvedata flexibilityVSAvoiddatabase schema complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates virtual copies of database tables through view definitions that can be dynamically modified without changing the underlying physical database structure. These virtual copies allow different machine learning algorithms to access data in different formats and structures simultaneously, resolving the contradiction between data flexibility and database complexity.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system dynamically generates and modifies view definitions based on the specific requirements of different machine learning algorithms. The view definitions can be created, modified, and deleted on-the-fly without requiring changes to the rigid database schema, enabling adaptive data access patterns.

Inventive Principle:
Principle #15Dynamics

2Adaptability or versatility

If data is extracted and transformed for each machine learning algorithm, then algorithm-specific requirements are met, but data preparation time increases

Engineering Contradiction:
Improvealgorithm compatibilityVSAvoiddata preparation time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent pre-defines multiple view definitions that correspond to different machine learning algorithm requirements. When a new algorithm needs to be trained, the appropriate pre-defined view can be selected and modified minimally, rather than performing complete data extraction and transformation from scratch.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The view mechanism provides a universal interface that can serve multiple machine learning algorithms simultaneously. A single view definition can be shared across multiple algorithms with different parameter requirements, reducing redundant data preparation work.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Quantity of substance

If raw data is stored without transformation, then storage efficiency is improved, but data cannot be directly used by machine learning algorithms requiring specific formats

Engineering Contradiction:
Improvestorage efficiencyVSAvoiddata usability
Core Design Contradiction:
Quantity of substanceVSEase of operation

Solution Approach 1:

The patent creates virtual copies of the raw data through view definitions that present the data in algorithm-ready formats without physically transforming or duplicating the underlying stored data. This maintains storage efficiency while providing usable data formats on-demand.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3948577B1Automated machine learning on the basis of stored data
Publication Date: 2026.05.06 BUNDESDRUCKEREI GMBH
  • EP3948577B1 patent drawingFigure 1
  • EP3948577B1 patent drawingFigure 2
  • EP3948577B1 patent drawingFigure 3

AI summary

The invention relates to a method for automated machine learning. A pre-trained learning module (120) and a database (104) are provided. An additional dataset (108) is received and stored in a first data model (106) of the database, and one or more tokens are (109) generated and assigned to one or more token types (111). The token assignments (113) are each classified as verified fact or preliminary assumption, and an index (112) in a second data model (110) is extended using the additional token, a pointer (115) to the additional stored data set, and the token assignments.