Mediation Parameter Model Adaptation for Limited Target Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Domain adaptation of recognition models is challenging when there is a limited amount of data and an inadequate calculation environment in the target domain, making it difficult to achieve good performance.
Innovation Solution
A model generation device and method that learns model parameters from multiple source domains and generates mediation parameter relevance information, allowing for the adjustment of target model parameters using evaluation data from the target domain to adapt the model to the target environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If additional learning is performed using target domain data, then model performance in target domain is improved, but requirement for sufficient learning data and calculation environment in target domain increases
Solution Approach 1:
The patent performs preliminary learning in source domains before target domain adaptation. Model parameters are pre-trained using abundant source domain data, and then fine-tuned with limited target domain data. This preliminary action in source domains prepares the model to achieve good performance even with limited target domain resources.
Solution Approach 2:
The patent introduces mediation parameters as an intermediary between source domain model parameters and target domain adaptation. These mediation parameters enable the model to adapt to target domain characteristics without requiring sufficient target domain learning data, acting as a bridge that transfers knowledge from source to target domains.
2Adaptability or versatility
If model parameters are corrected and interpolated using learning data from multiple domains, then domain adaptation capability is improved, but complexity of parameter adjustment increases
Solution Approach 1:
The patent introduces mediation parameters as an intermediary mechanism to simplify domain adaptation. Instead of directly correcting and interpolating model parameters from multiple domains, the mediation parameters serve as a controlled interface that manages the adaptation process, reducing the complexity of parameter adjustment while maintaining adaptability.
Solution Approach 2:
The patent systematically changes parameters from source domains to target domain by introducing mediation parameters. This structured parameter transformation approach converts the complex task of multi-domain parameter interpolation into a more manageable parameter change process guided by mediation parameters.
Data Source
AI summary
The model generation device generates model parameters corresponding to the model to be used and mediation parameter relevance information indicating the relevance between the model parameters of a plurality of source domains and the mediation parameters by using the learning data in the plurality of source domains. The model adjustment device generates target model parameters which correspond to the target domain and include the mediation parameters, based on the learned model parameters for each of the plurality of source domains and the mediation parameter relevance information. Then, the model adjustment device uses the evaluation data of the target domain to determine the mediation parameters included in the target model parameters.