Domain Adaptation Learning Apparatus for Low-Data Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge is to maintain recognition accuracy when there is an insufficient amount of training data for a new class in a specific domain, leading to decreased recognition performance.
Innovation Solution
A learning apparatus and method that includes units for extracting unique and common features from target and source domain data, calculating distances between these features, and optimizing parameter updates to enhance recognition accuracy across domains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If domain adaptation is applied to train an object recognition model using training data from different environments, then recognition accuracy for images from different environments is improved, but recognition accuracy for new classes with insufficient training data in a specific domain decreases
Solution Approach 1:
The patent segments the feature extraction process into two distinct pathways: a target domain feature extraction unit that processes only target domain data, and a common feature extraction unit that processes both source and target domain data. This segmentation allows the model to handle new classes with insufficient target domain data by leveraging common features from the source domain, while maintaining domain-specific accuracy for well-represented classes through target domain-specific features.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of a common feature extraction unit that acts as a mediator between source domain and target domain data. This intermediary extracts features that are common across both domains, enabling the model to compensate for insufficient training data in the target domain by utilizing knowledge transferred from the source domain through this intermediate feature representation.
2Quantity of substance
If training data for new classes is insufficient in a specific domain, then the model cannot learn adequate features for those classes, but using source domain data may introduce domain-specific biases
Solution Approach 1:
The patent applies local quality by making different parts of the feature extraction system serve different purposes: the target domain feature extraction unit focuses on capturing domain-specific characteristics for classes with sufficient data, while the common feature extraction unit captures generalizable features from both domains. This localized functional differentiation allows the model to utilize source domain data for new classes without introducing harmful domain-specific biases, as each extraction unit is optimized for its specific purpose.
Data Source
AI summary
The learning apparatus classifies target domain data into (N-c) classes based on unique features of the target domain data, classifies source domain data into N classes based on unique features of the source domain data, and classifies the target domain data and the source domain data into the N classes based on common features of the target domain data and the source domain data. Also, the learning apparatus calculates a first distance between the common features of the target domain data and the source domain data, and calculates a second distance between the unique features of the target domain data and the source domain data. Next, the learning apparatus updates parameters of a common feature extraction unit based on the first distance, and updates parameters of a target domain feature extraction unit and a source domain feature extraction unit based on the second distance.