Coupled Artificial Learning Units With Projection-Level Modulation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing artificial intelligence systems, particularly neural networks, are specialized for specific tasks and require extensive retraining for different applications, leading to high-dimensional spaces that hinder real-time reactions and efficient adaptability.

Innovation Solution

A system comprising coupled artificial learning units, where a first unit provides rapid categorization and influences a second, more complex unit through modulation functions and dropout methods, allowing for efficient and timely decision-making without complete retraining.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single specialized neural network is used for a specific task, then task performance is optimized, but adaptability to other domains is lost and complete retraining is required

Engineering Contradiction:
Improvetask performanceVSAvoiddomain adaptability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system segments the neural network into multiple specialized units (first artificial learning unit, second artificial learning unit, third artificial learning unit), each optimized for specific functions. This allows the system to maintain task-specific optimization while achieving domain adaptability through selective activation and composition of different units.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal system that can perform multiple domain-specific tasks by combining specialized learning units. The system achieves multi-functionality through the projection level that integrates outputs from different units, allowing adaptation to various domains without complete retraining.

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

2Adaptability or versatility

If a universally applicable AI system is designed to cover high-dimensional spaces, then adaptability is improved, but training and test data requirements increase exponentially

Engineering Contradiction:
Improveuniversal applicabilityVSAvoidtraining data volume
Core Design Contradiction:
Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

Instead of creating one large universal network requiring exponential data, the system segments functionality into multiple specialized units. Each unit can be trained independently on smaller, domain-specific datasets, avoiding the exponential data requirement while maintaining universal applicability through composition.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system employs a nested structure where specialized learning units are integrated within a larger universal framework. The projection level acts as the outer layer that coordinates specialized units, allowing the system to handle high-dimensional spaces through hierarchical organization rather than requiring exponentially more training data.

Inventive Principle:
Principle #7Nested doll (Nesting)

3Measurement precision

If a deeply complex neural network is used for comprehensive analysis, then analysis depth is improved, but real-time reaction speed decreases

Engineering Contradiction:
Improveanalysis depthVSAvoidreal-time reaction speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The system divides complex analysis tasks across multiple specialized learning units rather than using a single deep network. This segmentation allows parallel processing of different aspects of the input, maintaining analysis depth while improving real-time reaction speed through distributed computation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system applies partial action by activating only the necessary learning units based on the input type and task requirements, rather than always running the full deep network. This selective activation maintains analysis depth for complex tasks while achieving real-time performance for simpler tasks by using only the required subset of units.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3973457B1Coupling multiple artificially learning units with a projection level
Publication Date: 2025.10.01 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3973457B1 patent drawingFigure 1
  • EP3973457B1 patent drawingFigure 2a~2h
  • EP3973457B1 patent drawingFigure 3

AI summary

The invention relates to a method in a system consisting of at least one second and a third artificially learning unit, having the steps of: inputting first input values on at least one second artificially learning unit and obtaining output values based on the input values from the at least one second artificially learning unit; at least temporarily storing situation data, said situation data comprising first input values and/or second output values of the at least one second unit; using the situation data as input values for the third artificially learning unit, said third artificially learning unit generating third output values in response to the input values; and checking whether the second output values of the at least one second unit satisfy one or more specified conditions on the basis of the third output values. The invention additionally relates to a system for carrying out such a method.