Feature extraction by non parametric mutual information maximization
The Journal of Machine Learning Research
IEEE Transactions on Pattern Analysis and Machine Intelligence
IEEE Transactions on Robotics - Special issue on rehabilitation robotics
Co-training with relevant random subspaces
Neurocomputing
Convolutional Neural Networks for P300 Detection with Application to Brain-Computer Interfaces
IEEE Transactions on Pattern Analysis and Machine Intelligence
Input feature selection for classification problems
IEEE Transactions on Neural Networks
Estimating optimal feature subsets using efficient estimation of high-dimensional mutual information
IEEE Transactions on Neural Networks
Maximization of Mutual Information for Supervised Linear Feature Extraction
IEEE Transactions on Neural Networks
Using mutual information for selecting features in supervised neural net learning
IEEE Transactions on Neural Networks
Hi-index | 0.01 |
The detection of the presence of the P300 in the electroencephalogram (EEG) is a challenging issue in P300-based brain-computer interface (BCI). The P300-based developed in this work allows a subject to communicate 1 of 36 symbols presented on a 6x6 matrix. When a target symbol is seen by the subject, unique event-related potential (ERP) characterized by P300 is elicited. Thus, in P300 speller, accurate detection of such distinct ERP provides faster and reliable communication. In this paper, a subspace-based spatial filter was employed to enhance the detection of the presence of the P300 in the EEG. The subspace-based filter was designed by maximizing the ratio between the brain signals synchronized with the target stimulus and that with the nontarget stimulus. The processing method was evaluated offline and online on the data obtained from five subjects. The results of offline studies showed that the average accuracies of 97.5% and 90.5% were achieved in P300 detection and character recognition, respectively. The average communication rate achieved was 17.13bits/min with an average accuracy of 89.5% using 10 electrodes during online experiments.