Rule Learning from Knowledge Graphs Using Differentiable Operators

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for learning rules from knowledge graphs are limited in handling numeric data and do not effectively incorporate aggregate values or negations, which are crucial for predicting and completing relations between entities.

Innovation Solution

A computer-implemented method that derives aggregate values from numerical and non-numerical relations, encodes these into non-numerical relations, constructs differentiable operators, and uses a neural network to extract rules, allowing for the integration of aggregate values and negations into the rule learning process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If aggregate values are added as new entities in the form of numerical value to the data base, then the data base can store aggregate information, but it creates dense matrices that cannot be materialized

Engineering Contradiction:

Improveaggregate valuesVSAvoiddense matrix materialization

Core Design Contradiction:

Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent extracts aggregate values from numerical relations and derives non-numerical relations from them, separating the aggregate information handling from the original numerical data structure. This extraction allows the system to work with derived non-numerical relations instead of materializing dense matrices of aggregate values.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces derived non-numerical relations as an intermediary between numerical relations and the rule learning process. These derived relations act as a mediator that captures aggregate information without requiring direct materialization of dense matrices, enabling efficient processing through differentiable operators.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of manufacture

If existing methods are used for learning rules from knowledge graphs, then the process is simple, but they are limited in handling numeric data and aggregate values

Engineering Contradiction:

Improverule learning processVSAvoidhandling numeric data

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal framework that handles both numerical and non-numerical relations through the same rule learning process. By deriving non-numerical relations from aggregate values and constructing differentiable operators that work with both relation types, the system achieves multi-functionality while maintaining ease of use.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent changes the parameter representation by transforming numerical aggregate values into derived non-numerical relations. This parameter transformation allows the rule learning system to handle numeric data effectively while maintaining compatibility with existing logical rule frameworks through differentiable operators.

Inventive Principle:

Principle #35Parameter changes

3Loss of information

If aggregate values are derived and used in rule learning, then predictions can incorporate summary information, but the processing complexity increases

Engineering Contradiction:

Improvesummary over relationsVSAvoidprocessing complexity

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The patent performs preliminary actions by deriving aggregate values and creating derived non-numerical relations before the rule learning process. This preprocessing step captures summary information in advance, allowing the main rule learning algorithm to work with pre-computed relations rather than processing raw numerical data during training.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional mechanical aggregation operations with differentiable operators that can be integrated into neural network-based rule learning. This substitution allows gradient-based optimization to handle aggregate information processing, reducing processing complexity through automated differentiation rather than manual aggregation logic.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11699076B2Method and system for learning rules from a data base

Publication Date: 2023.07.11 ROBERT BOSCH GMBH

AI summary

A system and computer implemented method for learning rules from a data base including entities and relations between the entities, wherein an entity is either a constant or a numerical value, and a relation between a constant and a numerical value is a numerical relation and a relation between two constants is a non-numerical relation. The method includes: deriving aggregate values from said numerical and/or non-numerical relations; deriving non-numerical relations from said aggregate values; adding said derived non-numerical relations to the data base; constructing differentiable operators, wherein a differentiable operator refers to a non-numerical or a derived non-numerical relation of the data base, and extracting rules from said differentiable operators.