Dual-Encoder Learning for Cross-Domain Sensor Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing domain adaptation methods for estimating events from sensor data suffer from high introduction costs and information loss, leading to degraded estimation accuracy due to differences in data acquisition attributes.
Innovation Solution
A learning apparatus with two encoders and metadata identifiers is used to extract features and attributes from training data, performing adversarial training to separate domain-specific and common information, allowing for robust estimation across different data acquisition conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If domain adaptation methods are used to estimate events from sensor data across different domains, then the model can be applied to multiple data acquisition conditions, but the introduction cost increases and information loss occurs leading to degraded estimation accuracy
Solution Approach 1:
The patent segments the feature extraction process into two distinct encoders: a domain-specific encoder that captures characteristics unique to each data acquisition domain, and a common encoder that captures shared characteristics across domains. This segmentation allows the model to preserve domain-specific information while adapting to multiple domains, thereby maintaining estimation accuracy across different domains without requiring a complete redesign for each domain.
2Adaptability or versatility
If domain adaptation methods are used to estimate events from sensor data across different domains, then the model can be applied to multiple data acquisition conditions, but the introduction cost increases
Solution Approach 1:
The patent creates a universal framework where the common encoder and estimator components can be shared across multiple domains, while only the domain-specific encoder needs to be adapted for each new domain. This multi-functionality reduces the overall introduction cost when deploying the model to new domains, as the majority of the model architecture remains reusable and only requires minimal domain-specific adaptation.
3Adaptability or versatility
If domain adaptation methods are used to estimate events from sensor data across different domains, then the model can be applied to multiple data acquisition conditions, but information loss occurs leading to degraded estimation accuracy
Solution Approach 1:
The patent segments the feature extraction process into two distinct encoders: a domain-specific encoder that captures characteristics unique to each data acquisition domain, and a common encoder that captures shared characteristics across domains. This segmentation allows the model to preserve domain-specific information while adapting to multiple domains, thereby maintaining estimation accuracy across different domains without requiring a complete redesign for each domain.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A trained model is constructed whose introduction cost is relatively low and that is robust to the difference in the attribute regarding acquisition of data. A learning apparatus according to one aspect of the invention executes, with respect to each learning data set, a first training step of training a second encoder and a second metadata identifier such that the identification result by the second metadata identifier matches the metadata, a second training step of training encoders and an estimator such that the result of estimation performed by the estimator matches correct answer data, a third training step of training a first metadata identifier such that the result of identification performed by the first metadata identifier matches the metadata, and a fourth training step of training a first encoder such that the result of identification performed by the first metadata identifier does not match the metadata. The third training step and the fourth training step are alternatingly and repeatedly executed.