Generative Inter-Subject Transfer Learning for EEG Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing brain-computer interface (BCI) techniques face challenges due to the inherent deficiencies in the quality and quantity of EEG data, which are exacerbated by noise and external artifacts. Additionally, traditional domain adaptation methods require labeled target data, which can be difficult to obtain and resource-intensive.
Innovation Solution
A generative transfer learning apparatus and method that employs an unsupervised domain adaptation approach. This involves a data input unit, preprocessing unit, encoding units, generation unit, discrimination unit, and classification unit, which work together to remove outliers, augment data, and align the distribution of source and target data, enabling effective transfer learning without labeled target data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional domain adaptation methods are used, then classification performance can be improved, but labeled target data is required which increases resource consumption and complexity
Solution Approach 1:
The patent uses generative adversarial networks to create synthetic target domain data copies from source domain data. The generator creates artificial target domain samples that mimic the target domain distribution, allowing the system to train without actual labeled target data while maintaining classification performance
Solution Approach 2:
The system performs self-supervised learning where the generator and discriminator work together to create and verify synthetic data. The discriminator evaluates whether generated data resembles target domain data, enabling the system to automatically learn domain adaptation without external labeled target data
2Adaptability or versatility
If domain generalization is applied, then adaptability to new subjects is improved, but practical implementation becomes difficult
Solution Approach 1:
The patent introduces a generative adversarial network as an intermediary between source domain data and target domain classification. The generator acts as a mediator that transforms source data into synthetic target domain data, making domain adaptation practical while maintaining adaptability to new subjects
Solution Approach 2:
The system changes the distribution parameters of the data through generative adversarial training. By adjusting the generator's parameters to match target domain characteristics, the system achieves practical domain adaptation while maintaining the ability to generalize to new subjects
3Productivity
If shared distribution learning is performed, then transfer learning efficiency is improved, but target data integrity may be damaged
Solution Approach 1:
The patent extracts domain-specific features from source data and separates them from task-specific features. The generator extracts and transforms only the domain distribution characteristics while preserving target domain integrity, allowing efficient transfer learning without compromising target data quality
Solution Approach 2:
The system applies different processing quality to different aspects of data. The generator focuses on matching domain distribution locally while the classification model maintains target domain integrity locally, achieving both transfer learning efficiency and data quality preservation
Data Source
AI summary
A generative inter-subject transfer learning apparatus includes: a data input unit for receiving source data and target data; a preprocessing unit for removing outliers from the source data and augmenting data; a first encoding unit for generating a feature vector of the source data received from the preprocessing unit and a feature vector of the target data; a generation unit for generating transfer data by reconfiguring the feature vector of the source data, and trained so that a domain of the source data follows a domain of the target data; a second encoding unit for generating a feature vector of the transfer data; and a classification unit for classifying the feature vector of the transfer data generated by the second encoding unit, and trained so that a classification result of the transfer data matches a classification result of the source data.