Edge Cloud ML Agent for Resource-Adaptive Industrial Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited computing resources in edge clouds often make it impossible to execute complex machine learning operations, requiring additional resources from central clouds, which results in high latency and bandwidth consumption.
Innovation Solution
A machine learning agent identifies the state of an industrial process, selects and adapts a learning model's training algorithm to optimize resource usage within the edge cloud, allowing computations to be performed locally without additional resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If complex machine learning operations are executed in the edge cloud, then the latency is reduced and real-time processing is achieved, but the limited computing resources in the edge cloud are insufficient to support the computations
Solution Approach 1:
The patent applies dynamics by making the machine learning model adaptive and flexible in the edge cloud. The model can dynamically adjust its complexity and resource requirements based on available computing resources, enabling real-time processing without requiring fixed, over-provisioned infrastructure. This allows the system to optimize between model accuracy and resource consumption in real-time.
Solution Approach 2:
The patent utilizes parameter changes by modifying the machine learning model's parameters and configuration to suit the edge cloud's computing capabilities. By adjusting model parameters such as precision, complexity, and computation intensity, the system can execute complex operations within resource constraints while maintaining acceptable performance and low latency.
2Productivity
If computing resources are increased in the edge cloud to support complex machine learning operations, then the processing capability is improved, but the cost and complexity of the edge cloud infrastructure increases
Solution Approach 1:
The patent applies copying by using software-based virtualization and emulation to create virtual computing resources in the edge cloud. Instead of physically increasing hardware capacity, the system creates virtual copies of computing environments that can be dynamically allocated and managed, providing enhanced processing capability without proportional increases in physical infrastructure complexity.
Solution Approach 2:
The patent implements universality by designing a machine learning model that can perform multiple functions and adapt to different computational requirements within the same edge cloud infrastructure. The model can dynamically switch between different operation modes and complexity levels, allowing a single infrastructure to handle diverse workloads without requiring specialized hardware for each function.
3Quantity of substance
If machine learning operations are executed in the central cloud instead of the edge cloud, then sufficient computing resources are available, but the latency increases and bandwidth consumption increases
Solution Approach 1:
The patent applies segmentation by dividing the machine learning workload into segments that can be executed locally in the edge cloud. By segmenting the model into smaller, independently executable components, the system can process data locally without transmitting entire datasets to the central cloud, thereby reducing latency while utilizing distributed computing resources effectively.
4Measurement precision
If the machine learning model is optimized for accuracy, then the computational requirements increase, but the available computing resources in the edge cloud are exceeded
Solution Approach 1:
The patent utilizes parameter changes by dynamically adjusting the machine learning model's parameters such as precision, complexity, and computation intensity based on available computing resources. This allows the system to optimize between model accuracy and resource consumption, selecting appropriate parameter configurations that maintain acceptable accuracy while fitting within edge cloud resource constraints.
Solution Approach 2:
The patent applies dynamics by making the model adaptive, allowing it to dynamically adjust its accuracy and computational requirements based on real-time resource availability. The model can switch between different accuracy levels and complexity modes, enabling it to maintain high accuracy when resources are abundant and reduce accuracy requirements when resources are constrained.
Data Source
AI summary
Method and machine learning agent for executing machine learning on an industrial process by using computing resources in an edge cloud. A state of the industrial process is identified (2:1) and a learning model comprising a training algorithm for the machine learning is selected (2:2) based on the identified state. The training algorithm in the selected model is then adapted (2:4) so that the amount of available computing resources in the edge cloud is sufficient for computations in the training algorithm. The adapted training algorithm is finally applied (2:5) on data generated in the industrial process using computing resources in the edge cloud. Thereby, computing resources in the edge cloud can be used and no additional resources are needed, thus reducing latency and bandwidth consumption.