Neuromorphic Knowledge Graph Hardware for Online Triple Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems for training AI on knowledge graphs require extensive data extraction and processing, which is inefficient and limits the ability to perform online learning and inference directly on industrial devices, especially in dynamic and sparse data environments like industrial automation systems.

Innovation Solution

The development of neuromorphic hardware and methods that enable online learning and inference by using a probabilistic, sampling-based model for knowledge graphs, allowing for data-driven and model-driven learning modes without the need for negative training examples, and integrating graph embedding algorithms into neuromorphic processors for edge computing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing systems extract and process large quantities of raw data externally before training AI, then training accuracy can be improved, but processing time and system complexity increase significantly

Engineering Contradiction:

Improvetraining accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent combines data extraction, processing, and AI training functions into a single integrated neuromorphic system. The knowledge graph processor directly consumes triple statements from the knowledge graph and performs training in-place, eliminating the need for separate external data processing pipelines while maintaining training accuracy through specialized neuromorphic architectures.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system segments the knowledge graph into structured triple statements (subject-predicate-object) that can be directly processed by the neuromorphic hardware. This segmentation allows the system to work with structured data representations that are optimized for neuromorphic processing, avoiding the need to process large quantities of unstructured raw data externally.

Inventive Principle:

Principle #1Segmentation

2Productivity

If data processing is performed externally before deployment, then model training can be completed, but real-time online learning capability is lost

Engineering Contradiction:

Improvetraining completionVSAvoidonline learning capability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The neuromorphic knowledge graph processor is designed with dynamic reconfigurability, allowing it to switch between different processing modes and adapt to new data patterns in real-time. The system can continuously learn from incoming triple statements and update its models on-the-fly, enabling both training completion and real-time online learning capabilities simultaneously.

Inventive Principle:

Principle #15Dynamics

3Ease of operation

If classical statistical methods are used on symbolic graph data, then traditional machine learning approaches can be applied, but the symbolic nature of graph data prevents direct usage

Engineering Contradiction:

Improveapplicability of ML methodsVSAvoiddata preprocessing complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent replaces classical statistical methods and traditional machine learning approaches with neuromorphic computing methods specifically designed for symbolic graph data. The neuromorphic processor uses spiking neural networks and event-driven architectures that natively handle symbolic triple statements, eliminating the need for complex data preprocessing and conversion steps required by traditional methods.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

4Reliability

If graph embedding algorithms are implemented on traditional processors, then inference can be performed, but energy consumption and processing latency increase

Engineering Contradiction:

Improveinference capabilityVSAvoidenergy consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system changes the operational parameters of the processing system by using neuromorphic architectures with event-driven, sparse, and asynchronous processing modes. This allows graph embedding algorithms to run with significantly lower energy consumption compared to traditional processors, while maintaining inference capability through specialized hardware optimizations for knowledge graph processing.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20220230056A1Neuromorphic hardware for processing a knowledge graph represented by observed triple statements and method for training a learning component

Publication Date: 2022.07.21 SIEMENS AG

AI summary

Provided is neuromorphic hardware for processing a knowledge graph, with a learning component, having an input layer containing node embedding populations of neurons, with each node embedding populations representing an entity contained in the observed statements, and an output layer, containing output neurons configured for representing a likelihood for each possible triple statement, and modeling a probabilistic, sampling-based model derived from an energy function, wherein the observed statements have minimal energy, and with a control component, configured for switching the learning component into a data-driven learning mode, configured for training the component with a maximum likelihood learning algorithm minimizing energy in the probabilistic, sampling-based model, using only the observed statements, which are assigned low energy values, in which the learning component supports generation of triple statements, and into a model-driven learning mode, configured for training the component, with the learning component learning to assign high energy values.