Model Generation with Local Identification Models for Diverse Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods face challenges in constructing a generation model that can generate a wide variety of data due to limitations in collecting sufficient and diverse learning data across multiple sites, leading to inaccurate visual inspection and inference, while also posing issues with data confidentiality and high communication and calculation costs.
Innovation Solution
A model generation system that transmits data generated by a generation model to identification models for identification, collects results of these identifications, and trains the generation model using these results to degrade the performance of specific identification models, thereby constructing a model capable of generating diverse data without directly using local learning data, thus ensuring confidentiality and reducing communication and calculation costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a sufficient amount of inference learning data is collected to improve visual inspection accuracy, then the accuracy of visual inspection is improved, but the cost of collecting data increases
Solution Approach 1:
The patent uses a generation model to create synthetic inference learning data that copies the statistical distribution and characteristics of real learning data. This allows sufficient training data to be generated without the high costs associated with collecting large amounts of real data, while maintaining the accuracy needed for visual inspection
2Adaptability or versatility
If learning data from multiple sites is gathered to construct a generation model, then the diversity of generated data is improved, but data confidentiality is compromised
Solution Approach 1:
The patent segments the learning process by training separate identification models at each site using only local data. These distributed models then collaborate through federated learning to collectively train the generation model, achieving data diversity without centralizing sensitive data and thus preserving confidentiality
Solution Approach 2:
The patent introduces identification models as intermediaries that process and validate data contributions from multiple sites. These intermediaries enable the system to leverage diverse data sources while maintaining security through controlled access and verification mechanisms
3Adaptability or versatility
If learning data from multiple sites is gathered to construct a generation model, then the diversity of generated data is improved, but communication and calculation costs increase
Solution Approach 1:
The patent implements self-service through federated learning, where each site independently trains its own identification model using local data. This eliminates the need to transfer large amounts of data across the network, significantly reducing communication costs while still achieving diverse model training
Solution Approach 2:
The training process is segmented into distributed local training operations and coordinated model aggregation. This segmentation allows computation to be performed locally where data resides, minimizing data transmission requirements and reducing overall communication and calculation costs
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
Provided is a technique for constructing a generation model that can generate various data. A model generation apparatus according to one aspect of the invention includes: a generating unit that generates data using a generation model; a transmitting unit that transmits the generated data to a plurality of trained identification models that each have acquired, by machine learning using local learning data, a capability of identifying whether or not given data is the local learning data, and causes the identification models to perform an identification on the data; a receiving unit that receives results of identification with respect to the transmitted data executed by the identification models; and a learning processing unit that trains the generation model to generate data that causes identification performance of at least one of the plurality of identification models to be degraded, by performing machine learning using the received results of identification.