Bayesian optimization of BCI parameters

An important factor in custom designing a brain computer interface, BCI, is the estimation of the values of its parameters. This paper proposes a fully automatic algorithm that uses Bayesian optimization to tune the hyper-parameters of a synchronous BCI. The algorithm finds a large number of possible sets of values for the hyper-parameters. Each set is then used to train the classifier and the results over the possible sets of hyper-parameter values are aggregated. In this paper we consider a simple motor imagery based BCI with two parameters: the EEG frequency bands and the time intervals from which the features are extracted. We use the linear discriminant analysis classifier and aggregate all results using multi-response linear regression. Experiments using the BCI competition III dataset 3b show that our proposed method results in considerable improvement in the accuracy of a BCI. The average accuracy of our method was 2.6% better than the best results obtained by existing methods.

[1]  K.-R. Muller,et al.  Optimizing Spatial filters for Robust EEG Single-Trial Analysis , 2008, IEEE Signal Processing Magazine.

[2]  G. Pfurtscheller,et al.  Brain-Computer Interfaces for Communication and Control. , 2011, Communications of the ACM.

[3]  Nando de Freitas,et al.  A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning , 2010, ArXiv.

[4]  Ian H. Witten,et al.  Issues in Stacked Generalization , 2011, J. Artif. Intell. Res..

[5]  Anatole Lécuyer,et al.  Exploring two novel features for EEG-based brain-computer interfaces: Multifractal cumulants and predictive complexity , 2010, Neurocomputing.

[6]  Jinglu Hu,et al.  A novel frequency band selection method for Common Spatial Pattern in Motor Imagery based Brain Computer Interface , 2010, The 2010 International Joint Conference on Neural Networks (IJCNN).

[7]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.

[8]  J. Wolpaw,et al.  Mu and Beta Rhythm Topographies During Motor Imagery and Actual Movements , 2004, Brain Topography.

[9]  Dennis J. McFarland,et al.  Brain–computer interfaces for communication and control , 2002, Clinical Neurophysiology.

[10]  Anatole Lécuyer,et al.  Comparative study of band-power extraction techniques for Motor Imagery classification , 2011, 2011 IEEE Symposium on Computational Intelligence, Cognitive Algorithms, Mind, and Brain (CCMB).

[11]  Clemens Brunner,et al.  Automatic frequency band selection for BCIs with ERDS difference maps , 2011 .

[12]  F. L. D. Silva,et al.  Event-related EEG/MEG synchronization and desynchronization: basic principles , 1999, Clinical Neurophysiology.

[13]  Bernard Zenko,et al.  Is Combining Classifiers with Stacking Better than Selecting the Best One? , 2004, Machine Learning.

[14]  Anatole Lécuyer,et al.  Classifying EEG for brain computer interfaces using Gaussian processes , 2008, Pattern Recognit. Lett..

[15]  G. Pfurtscheller,et al.  The BCI competition III: validating alternative approaches to actual BCI problems , 2006, IEEE Transactions on Neural Systems and Rehabilitation Engineering.

[16]  T.M. McGinnity,et al.  Comparative Analysis of Spectral Approaches to Feature Extraction for EEG-Based Motor Imagery Classification , 2008, IEEE Transactions on Neural Systems and Rehabilitation Engineering.

[17]  Rabab K Ward,et al.  A survey of signal processing algorithms in brain–computer interfaces based on electrical brain signals , 2007, Journal of neural engineering.

[18]  Francesca Bovolo,et al.  Automatic selection of frequency and time intervals for classification of EEG signals , 2007 .