Increased accuracy in attention detection with inter-subject classification
A deep convolution neural network (CNN)-based framework for the classification of EEG into attentive/non-attentive mental states with the applications in cognitive BCI, game-based BCI and neuro-rehabilitation was developed. This framework significantly improves the attention detection accuracy with inter-subject classification.
In the adoption of deep learning on electroencephalogram (EEG)-based BCI, the loss of information is a critical problem. In our proposed Deep CNN framework, it does not suffer such losses from the transferences of learned knowledge to new subject. This is because it learns from raw EEG data with the least amount of pre-processing which in turn eliminates the extensive computational load of time-consuming data preparation and feature extraction.
Existing Fast Fourier Transform Support Vector Machine (FFT-SVM) and Linear Discriminant Analysis (LDA) methods used for attention detection classifications were taken as a baseline for comparison.
Our proposed Deep CNN methods have obtained an average classification accuracy of 79.94% which significantly surpasses existing methods by 10%.
This framework is beneficial in the application of attention-based BCI systems and extended to other types of EEG-based BCIs where the system can learn from raw EEG and successfully transfer the learned knowledge to a new target subject.
SCADAWall has the potential to achieve more security features, such as to prevent critical states or anomaly due to safety concern. With its CPI technology, it can maintain real-time communication without sacrificing network performance. This CPI technology may be further improved by testing against more SCADA protocols whose complexity in pay-load structure is higher than Modbus.
The A*STAR-affiliated researchers contributing to this research are from the Healthcare team of Institute for Infocomm Research.
Paper can be found in:
Journal of Neural Engineering (JNE), Inter-subject transfer learning with end-to-end deep convolutional neural network for EEG-based BCI, Fatemeh Fahimi, Zhuo Zhang, Wooi Boon Goh, Tih-Shih Lee, Kai Keng Ang, Cuntai Guan,March 2019