Low-Code ML Data Stream Analysis via Edge Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems require extensive development processes, making them impractical for smaller-scale or time-constrained applications, where rapid deployment and customization of data stream analysis are needed.
Innovation Solution
A cloud-based system that uses model templates and graphical user interfaces to train and deploy machine learning models, allowing users to select and customize models without coding, and deploys these models to edge datacenters for efficient data stream analysis, reducing latency and enabling low-code/no-code implementation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a conventional machine learning system is developed from scratch, then the system can be customized to specific data analysis needs, but the development process becomes lengthy and complex
Solution Approach 1:
The patent applies preliminary action by pre-training multiple machine learning models on different aspects of data stream analysis before deployment. These pre-trained models are stored in memory and can be rapidly selected and deployed without requiring time-consuming training from scratch, thus reducing development time while maintaining customization capability through model selection and combination
Solution Approach 2:
The system implements universality by creating a unified machine learning system that can perform multiple different data analysis functions through a single platform. The system can select from various pre-trained models and combine them to handle different analysis needs (anomaly detection, pattern recognition, prediction, etc.), eliminating the need to develop separate systems for each function
2Measurement precision
If multiple machine learning models are combined for comprehensive data analysis, then the analysis capability is improved, but the computational resources and processing time increase
Solution Approach 1:
The system applies partial action by selectively activating only the necessary pre-trained models based on the specific analysis requirements and data characteristics. Rather than running all available models simultaneously, the system chooses the minimal subset needed to achieve the desired analysis accuracy, thus reducing computational resource consumption while maintaining sufficient precision
Solution Approach 2:
By pre-training models beforehand and storing them in memory, the system avoids the excessive computational resources that would be required for real-time training of multiple models. The pre-trained models are ready for immediate deployment, eliminating the energy-intensive training phase while maintaining the capability to combine multiple models for comprehensive analysis
3Speed
If machine learning models are deployed to edge datacenters for real-time analysis, then the response latency is reduced, but the deployment complexity increases
Solution Approach 1:
The system extracts the machine learning models from centralized cloud environments and deploys them to edge datacenters closer to the data sources. This extraction enables real-time analysis at the edge with reduced latency, while the standardized model format and automated deployment process manage the complexity of distributed deployment
4Productivity
If a library of pre-trained models is maintained for rapid deployment, then the deployment speed is improved, but the storage requirements and model selection complexity increase
Solution Approach 1:
The system achieves universality by designing pre-trained models that can handle multiple data types and analysis scenarios. This multi-functionality reduces the total number of models needed in the library, as each model can be applied to various situations, thus decreasing storage requirements while maintaining rapid deployment capability through model selection
Data Source
AI summary
The present application relates to developing and deploying machine learning analysis systems using a low-code or no-code approach. A cloud service is configured to receive a first data stream from a sensor device and train a machine-learning model to recognize selected elements of the first data stream that are selected from a package of template models via a graphical user interface. The cloud service deploys the machine-learning model to an edge datacenter configured to receive a second data stream via a network connection. The edge datacenter locally interrogates the second data stream based on the machine-learning model to generate an element set including the selected elements. A logic service may receive a selection of one or more properties of the element set and one or more logical operators via a graphical user interface to generate user-configured logical rules. The logic service may the user-configured logical rules to the element set.