Industrial Automation ML Cohorting for Secure Cross-Site Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing industrial automation systems face challenges in developing robust machine learning (ML) models due to limited data availability and the need to share sensitive data across site boundaries, which increases communication overhead and security risks.
Innovation Solution
An industrial automation system that categorizes process components into cohorts based on cohorting criteria, allowing for the sharing of ML model parameters and data within neighboring cohort groups, thereby enhancing model robustness without exposing sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data from multiple sites are shared to train ML models, then model robustness is improved, but communication overhead and storage cost increase
Solution Approach 1:
The patent extracts and shares only the essential ML model parameters (weights, biases, gradients) rather than the complete raw training data. This extraction approach allows multiple sites to contribute to model training while minimizing the volume of data transmitted across the network, thereby improving model robustness without proportionally increasing communication overhead.
Solution Approach 2:
The patent segments the ML training process into distributed federated learning tasks where each site trains locally and shares parameter updates. This segmentation allows parallel processing across multiple sites, improving model robustness through diverse data sources while reducing the need for centralized data aggregation and the associated communication costs.
2Reliability
If data from multiple sites are shared to train ML models, then model robustness is improved, but storage cost increases
Solution Approach 1:
The patent extracts only the necessary ML model parameters (weights, biases, gradients) from the training process and shares these compact representations across sites. This extraction eliminates the need to store and transmit large volumes of raw training data, thereby improving model robustness through multi-site collaboration while keeping storage requirements minimal.
3Object-affected harmful factors
If ML models are trained on-site using local data, then data security is maintained, but model robustness deteriorates due to limited data availability
Solution Approach 1:
The patent introduces an intermediary federated learning framework that enables secure collaboration between sites. Each site keeps its raw data locally (maintaining security), while the intermediary mechanism facilitates the exchange of ML model parameters. This intermediary layer allows the system to achieve model robustness through diverse data sources without compromising data security, as no raw sensitive data leaves the local sites.
Solution Approach 2:
The patent segments the training data access into local-only raw data and shared ML parameters. This segmentation allows each site to maintain its data security boundaries while still benefiting from multi-site collaboration through the shared model parameters, thereby achieving both data security and model robustness simultaneously.
4Reliability
If complete training data are shared across sites, then model robustness is improved, but communication overhead and security risks increase
Solution Approach 1:
The patent extracts only the essential ML model parameters (weights, biases, gradients) from the complete training data and shares these extracted representations across sites. This extraction approach achieves model robustness through multi-site collaboration while dramatically reducing the communication overhead compared to sharing complete datasets, as model parameters are orders of magnitude smaller than raw training data.
Data Source
AI summary
An industrial automation system comprises multiple process components, each categorizable into a cohort corresponding to a cohorting criterion. Some process components are configured to perform a machine learning (ML) process. A process component hosts at least a part of an ML model per cohort and communicates the ML model parameters among the multiple process components. The system assigns one or more of the process components to one of the cohorts according to the cohorting criterion; attributes the ML model parameters of a process component in a selected one of the cohorts to the ML model belonging to the selected cohort; determines a proximity value of each pair of cohorts; assigns a pair of cohorts to a respective neighboring cohort group when the proximity value meets a predetermined proximity criterion; and shares the ML model related data between process components belonging to the same neighboring cohort group.