Edge Knowledge Graph Learning for Low-Latency Industrial Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for training AI methods on knowledge graph data require extensive offline processing and are not capable of continuous learning at the edge, leading to latency and inefficiencies in industrial applications where data is generated.
Innovation Solution
An industrial device and method that integrates learning and inference capabilities within the device, using neuromorphic hardware and algorithms like RESCAL, TransE, or Graph Convolutional Neural Networks to process knowledge graph data directly on edge devices, eliminating the need for external data processing and enabling continuous learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing systems use offline processing for training AI methods on knowledge graph data, then processing accuracy can be maintained, but latency and response time worsen due to the inability to perform continuous learning at the edge
Solution Approach 1:
The system segments the AI processing functionality into distributed edge devices that can independently perform learning and inference operations. Each edge device contains local processing capabilities including knowledge graph storage, learning algorithms, and inference engines, allowing data to be processed at the source rather than centralized offline, thus reducing latency while maintaining accuracy through distributed intelligent processing
Solution Approach 2:
The system performs preliminary actions by pre-loading knowledge graphs and training models directly onto edge devices before deployment. This allows the edge devices to immediately process incoming data without waiting for offline processing cycles, enabling continuous learning and real-time inference while maintaining the accuracy benefits of pre-trained models
2Device complexity
If data is processed externally rather than locally, then device complexity can be reduced, but productivity worsens due to the need for continuous data extraction and transmission
Solution Approach 1:
The edge devices are equipped with self-service capabilities including local knowledge graph storage, embedded learning algorithms, and autonomous inference engines. These devices can independently extract, process, and learn from data without requiring external processing infrastructure, thereby improving productivity through continuous local processing while managing complexity through integrated but modular architecture components
3Speed
If continuous learning is implemented at the edge, then responsiveness improves, but device complexity increases due to the integration of learning and inference capabilities
Solution Approach 1:
The learning and inference system is segmented into modular functional components including data ingestion modules, knowledge graph storage modules, learning algorithm modules, and inference engine modules. Each module can be independently configured and optimized, allowing edge devices to achieve rapid responsiveness through localized processing while managing complexity through clear functional separation and standardized interfaces
Solution Approach 2:
The edge devices are designed with multi-functional capabilities that combine data processing, knowledge graph management, learning, and inference in a single platform. This universal design allows the same device to perform multiple functions locally, improving responsiveness by eliminating data transmission delays while managing complexity through integrated but modular architecture that avoids redundant components
Data Source
AI summary
Provided is an industrial device for building and/or processing a knowledge graph, with at least one sensor and/or at least one data source configured for providing raw data, with an ETL component, configured for converting the raw data into triple statements, using mapping rules, with a triple store, storing the triple statements as a dynamically changing knowledge graph (with a learning component, configured for processing the triple statements in a learning mode, and for performing an inference in an inference mode, and with a control component, configured for switching between different modes of operation of the learning component.