2021/02/11 by Thomas Strypsteen, Strypsteen, Thomas, Alexander Bertrand +1 · 1 citation
Computer Science · Engineering · Neuroscience · #Advanced Memory and Neural Computing #Blind Source Separation Techniques #EEG and Brain-Computer Interfaces #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2102.09050
openalex publication_date 2021/02/11 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Many electroencephalography (EEG) applications rely on channel selection\nmethods to remove the least informative channels, e.g., to reduce the amount of\nelectrodes to be mounted, to decrease the computational load, or to reduce\noverfitting effects and improve performance. Wrapper-based channel selection\nmethods aim to match the channel selection step to the target model, yet they\nrequire to re-train the model multiple times on different candidate channel\nsubsets, which often leads to an unacceptably high computational cost,\nespecially when said model is a (deep) neural network. To alleviate this, we\npropose a framework to embed the EEG channel selection in the neural network\nitself to jointly learn the network weights and optimal channels in an\nend-to-end manner by traditional backpropagation algorithms. We deal with the\ndiscrete nature of this new optimization problem by employing continuous\nrelaxations of the discrete channel selection parameters based on the\nGumbel-softmax trick. We also propose a regularization method that discourages\nselecting channels more than once. This generic approach is evaluated on two\ndifferent EEG tasks: motor imagery brain-computer interfaces and auditory\nattention decoding. The results demonstrate that our framework is generally\napplicable, while being competitive with state-of-the art EEG channel selection\nmethods, tailored to these tasks.\n