Personalized State Estimation Without Per-User Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional state estimation systems require retraining of complex neural network models for each user, leading to significant time consumption and storage needs due to individual optimization, making it inefficient for large user bases.
Innovation Solution
A state estimation device that adds personalized optimization information to a parent model using detection data, allowing for user-specific estimation without retraining, by incorporating personalized optimization information to modify the model's output or layers, enabling accurate user-specific state estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the state estimation model is retrained for each user to optimize estimation accuracy, then the estimation accuracy for each user is improved, but the time consumption and storage requirements increase significantly
Solution Approach 1:
The patent divides the model optimization process into two segments: a general parent model that is trained once with diverse training data, and user-specific personalized optimization information that is computed individually for each user. This segmentation allows the system to avoid retraining the entire model for each user while still achieving personalized estimation accuracy.
Solution Approach 2:
Instead of creating entirely new models for each user through retraining, the patent uses the pre-trained parent model as a template and generates personalized versions by incorporating user-specific optimization information. This copying approach preserves the general knowledge in the parent model while adding user-specific adaptations without requiring full retraining.
2Measurement precision
If the state estimation model is retrained for each user to optimize estimation accuracy, then the estimation accuracy for each user is improved, but the storage requirements increase significantly
Solution Approach 1:
The patent extracts the essential user-specific information from the complete retraining process and represents it as compact personalized optimization information. This extracted information captures the necessary user adaptations in a condensed format that requires minimal storage space compared to storing full user-specific models.
Solution Approach 2:
The patent changes the representation of user-specific model parameters from full model retraining to optimized parameter adjustments. By representing user-specific adaptations as parameter changes or optimization information rather than complete models, the storage requirements are significantly reduced while maintaining estimation accuracy.
3Adaptability or versatility
If complex neural network models are used to improve estimation capability, then the estimation capability is improved, but the model size and retraining time increase
Solution Approach 1:
The patent creates a universal parent model that can serve multiple users through a single training process. This universal model is designed to handle diverse user data and can be adapted to different users through personalized optimization information, eliminating the need for separate complex models for each user while maintaining high estimation capability.
Data Source
AI summary
A state estimation device includes an addition unit configured to add personalized optimization information to a parent model for estimating a state of a person by inputting detection data detected from the person, the personalized optimization information being used for making estimation, which is performed using the parent model, suitable for a user; and a state estimation unit configured to estimate a state of the user by inputting detection data detected from the user, into the parent model to which the personalized optimization information added.