PROJECT TITLE :

Improving EEG Decoding via Clustering-Based Multitask Feature Learning

ABSTRACT:

Due to the significantly low signal-to-noise ratio of EEG collected at the brain scalp, one of the most difficult steps in the process of developing a brain–computer interface (BCI) is accurate electroencephalogram (EEG) pattern decoding for particular mental tasks. This is one of the key steps for the development of a BCI. The use of Machine Learning as a strategy to enhance EEG pattern optimization and thereby improve decoding accuracy is an exciting prospect. However, the currently available algorithms do not effectively explore the underlying data structure that can effectively capture the true EEG sample distribution, and as a result, they can only produce a suboptimal level of decoding accuracy. We propose a clustering-based multitask feature learning algorithm for improved EEG pattern decoding as a means of elucidating the intrinsic distribution structure of EEG data. This will allow us to uncover the EEG data. To be more specific, we use affinity propagation-based Clustering to investigate the subclasses (also known as clusters) that exist within each of the initial classes. After doing so, we give each subclass a distinct label by employing a one-versus-all encoding strategy. In order to jointly optimize the EEG pattern features from the uncovered subclasses, we devise a novel multitask learning algorithm using the encoded label matrix as our starting point. This algorithm takes advantage of the subclass relationship. After that, we train a linear support vector machine for EEG pattern decoding using the features that were optimized earlier. Extensive experimental studies are carried out on three different EEG data sets in order to validate the efficacy of our algorithm in comparison to other methods that are considered to be state-of-the-art. The improved experimental results demonstrate the outstanding superiority of our algorithm, suggesting that it has prominent performance for EEG pattern decoding when applied to BCI applications.