Domain Adaptation for Facility State Classification
Solution Overview
Problem
Existing technologies face challenges in effectively transferring learning between domains, particularly in classifying the quality of state data across different domains without sufficient labeled data in the target domain.
Innovation Solution
An apparatus and method that acquire target and source state data, generate a conversion rule to transform data into a common space, and create a classification model using labeled data in the common space to classify the quality of state data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transfer learning is applied to classify state data quality in target domain using source domain data, then classification accuracy can be improved with limited target domain data, but information loss occurs during domain transformation
Solution Approach 1:
The patent introduces a common space as an intermediary representation between source domain and target domain. State data from both domains are transformed into this common space through domain transformation, allowing the classification model to learn from source domain data while applying to target domain data without direct domain-to-domain mapping that would cause information loss.
Solution Approach 2:
The patent transforms state data from different domains by changing their parameter representations into a common space. This involves adjusting data parameters and distributions to be compatible across domains while preserving the essential quality information needed for classification.
2Adaptability or versatility
If domain transformation is used to create common space for classification, then model generalization across domains is improved, but data distribution differences between domains cause classification challenges
Solution Approach 1:
The patent addresses data distribution differences by transforming parameters of state data from source and target domains into a common parameter space. This parameter transformation aligns the distributions while maintaining the discriminative quality information, enabling reliable classification despite original domain differences.
Solution Approach 2:
The common space serves as a universal representation that can accommodate state data from multiple domains simultaneously. This universal space allows a single classification model to function reliably across different domains by processing all data through the same transformed representation.
Data Source
AI summary
An apparatus is provided that includes an acquisition unit for acquiring target state data that are the state data in the target domain and source state data that are state data in a source domain and to which a label indicating a quality of a state is added; a conversion generation unit for generating a conversion rule of the target state data and the source state data into data of a common space based on distribution of the target state data and the source state data; a model generation unit for generating classification model for classifying a quality of a state indicated by the target state data in the common space using data of the common space to which the label is added; and an evaluation unit for evaluating the classification model using the source state data converted into data of the common space.