Federated Learning Abnormality Detection via Pseudo Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection systems using federated learning face challenges in achieving both sufficient data confidentiality and high abnormality detection accuracy, particularly when data distribution across devices is uneven.
Innovation Solution
The system employs a configuration that includes initial model training, device ID generation, latent variable generation, pseudo data generation, and shared model training using a conditional generative model, ensuring confidentiality and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If federated learning is used to provide abnormality detection service to multiple business operators, then data confidentiality is improved, but abnormality detection accuracy deteriorates due to different data distributions across devices
Solution Approach 1:
The patent generates pseudo-data by copying and transforming real data through a conditional generative model. The pseudo-data preserves the statistical characteristics and distribution patterns of the original data while being sufficiently different to prevent direct exposure of actual data. This allows the central processing device to train a shared model using pseudo-data from multiple devices without accessing the real data, thus maintaining confidentiality while enabling accurate abnormality detection through the learned patterns.
Solution Approach 2:
The patent changes the parameters of the data by using a conditional generative model to transform real data into pseudo-data. The conditional aspect allows the model to generate pseudo-data with specific characteristics while maintaining the underlying distribution patterns. This parameter transformation enables the system to work with modified data that preserves essential patterns for accurate detection while preventing direct access to original data.
2Loss of information
If machine learning models are trained separately on data from each abnormality detection device, then data confidentiality is maintained, but model integration becomes difficult and abnormality detection accuracy decreases
Solution Approach 1:
Instead of integrating complex models directly, the patent creates pseudo-data copies that capture the essential patterns from each device's data distribution. The central processing device trains a single shared model on this pseudo-data, eliminating the need for complex model integration while maintaining the benefits of device-specific data patterns. This approach simplifies the system architecture significantly.
Solution Approach 2:
The pseudo-data acts as an intermediary between the device-specific data and the shared model. Rather than directly integrating models trained on real data from multiple devices, the pseudo-data serves as a mediator that preserves the distribution patterns while enabling centralized model training. This intermediary approach decouples the data from the model integration process.
3Adaptability or versatility
If the same abnormality detection system is provided to multiple business operators, then service scalability is improved, but data exposure to other operators increases
Solution Approach 1:
The patent uses pseudo-data that copies the statistical characteristics and distribution patterns of real data without containing actual sensitive information. This allows the system to serve multiple business operators with the same abnormality detection system while the pseudo-data ensures that no real data from one operator is exposed to others. The conditional generative model generates pseudo-data that maintains patterns necessary for detection while preventing data exposure.
Data Source
AI summary
An abnormality detection system detects an abnormality in equipment using a machine learning model constructed by federated learning. The system comprises multiple computers that further include an initial model training unit configured to construct an initial training model based on input data, a device ID generation unit configured to generate a device ID for uniquely identifying each computer, a latent variable generation unit configured to generate a latent variable to be input to a model, a pseudo data generation unit configured to generate pseudo data for each of the plurality of computers, and a shared model training unit configured to construct a shared model which is a conditional generative model based on the pseudo data. These computers also include multiple abnormality detection devices configured to detect an abnormality in equipment by calculating an abnormality degree of the equipment based on the input data.