Federated Model Integration for Low-Cost Cross-Site Inspection AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods face challenges in constructing a trained learning model with higher capability due to limitations in collecting large volumes of valuable learning data quickly, which is exacerbated by high communication and calculation costs, especially when individual sites operate independently and have varying data structures and models.
Innovation Solution
A model integration apparatus that collects trained learning models with a common structure and integrates their results within a specified range, reducing communication and calculation costs while allowing for different model structures, thereby enhancing the capability of the integrated model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If learning data from multiple sites is gathered to improve inspection capability, then the inspection capability is improved, but communication cost and calculation cost increase enormously
Solution Approach 1:
The patent segments the learning process by allowing each site to independently train its own learning model using local data, rather than centralizing all data in one location. This division reduces communication costs while still enabling capability improvement through model integration.
Solution Approach 2:
The patent introduces an integration model as an intermediary that combines the learning models from multiple sites. Instead of directly gathering and processing all raw data centrally, the system uses these integrated models as mediators to achieve improved inspection capability with reduced communication overhead.
2Reliability
If learning data is gathered from many sites to construct a trained learning model, then the inspection capability is improved, but calculation processing time lengthens and memory shortage occurs
Solution Approach 1:
The training process is segmented across multiple sites, with each site independently training its own model. This parallel processing approach reduces overall calculation time compared to centralized training, while the integration model combines these distributed results to achieve improved inspection capability.
Solution Approach 2:
Each site performs partial training action on its local data subset rather than all sites processing the complete dataset. This division of labor reduces the computational burden on any single system and decreases overall processing time while still achieving comprehensive model training through integration.
3Adaptability or versatility
If learning models with different structures are integrated, then adaptability to different sites is improved, but device complexity increases
Solution Approach 1:
The integration model serves as a universal structure that can accommodate and combine learning models with different structures from various sites. This multi-functional approach allows the system to handle diverse local models while maintaining a unified integration framework, balancing adaptability with manageable complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is a technique for constructing a trained learning model having a higher capability. A model integration apparatus according to one aspect of the invention includes a model collecting unit that collects trained learning models from a plurality of learning apparatuses, an integration processing unit that executes integration processing of integrating the results of machine learning reflected in an integration range set in the common portion, with respect to the trained learning models, and a model updating unit that transmits a result of the integration processing to the learning apparatuses, and update the trained learning models retained by the learning apparatuses by causing the learning apparatuses to each apply the result of the integration processing to the integration range in the trained learning model.