Title: Deep Feature Mining via Encoder-Decoder Networks for EEG classification Tasks
Presenter: Xiaomin Li
Advisor: Dr. Vangelis Metsis
Date/Time : Friday, May 7th @11:45 a.m. – 12:30 p.m.
Zoom Link: https://txstate.zoom.us/j/93246047075
Deep learning methods have shown essential impacts on building practical Electroencephalography (EEG) based Brain-Computer-Interface (BCI). However, due to its multiform collecting objectives, equipment settings, EEG recordings from each dataset can be varied in many ways. Re-using such heterogeneous EEG data to train a general feature extraction deep learning model is not as easy as other deep learning successful cases, such as ResNet for computer vision, Transformer for natural language processing, etc. In this study, we develop an encoder-decoder network to extract common features that co-exist on EEG recordings. We then transform feature maps obtained from different datasets to the same shape and train general deep EEG models in both supervised and unsupervised ways. In the end, we conduct comprehensive experiments to illustrate the benefits gained from encoder-decoder networks for transferring such general models to other EEG classification tasks.
Deadline: May 8, 2021, midnight