Routing Mechanisms in Convolutional Neural Networks for EEG Decoding

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Convolutional neural networks (CNNs) are becoming increasingly popular tools for decoding electroencephalographic (EEG) data. They alleviate the need for domain expertise in a typical decoding pipeline by being able to learn feature representations from raw input. However, the impact of new deep learning research ideas on EEG decoding is yet to be fully explored. This bachelor thesis introduces the idea of routing mechanisms in CNNs for EEG decoding. We present two mechanisms: routing through optimization and routing through attention. Each mechanism is implemented within the first layer of a CNN specifically designed to handle EEG data. Through our experimental setup, we aim to answer the following question: What is the impact of each routing mechanism on EEG decoding accuracies? To do so, we look at our models’ learning process and test for statistically significant differences in performance between our networks and a baseline implementing regular convolution. Our experimental results show that both mechanisms function effectively and our models reach at least 89% decoding accuracy. However, no statistically significant impact is found for either mechanism. The work here presented sets a framework that future research can build upon in order to create an efficient flow of information through the network.
Faculteit der Sociale Wetenschappen