ETRI Geriatric Depression Detection based on Unsupervised Learning

ETRI 클러스터 사업 Geriatric Depression Detection based on Unsupervised Learning 데이터 사이언스 연구실 김선영 2020 -08 -18

Reference Paper Deep Convolution Neural Network and Autoencoders. Based Unsupervised Feature Learning of EEG Signals IEEE Access (2018). The deep convolution network and autoencoders-based model, named as AE-CDNN, is constructed in order to perform unsupervised feature learning from EEG in epilepsy. - Extract features by AE-CDNN model and classify the features based on two public EEG data sets.

The model mainly has two stages: 1) Encoder Stage - Sample input, convolution layer, pooling layer(down sampling layer), and the feature coding 2) Decoder Stage - Feature coding input, full connection layer, reshape operation, deconvolution layer, upsampling layer and the reconstruction samples.

Point reduction 1초당 64개 1초당 8개 30분간 입력된 BVP 신호 # of raw signal points = 115, 200 = 30*60*64 1초당 32개 1초당 16개 1초당 4개 1초당 2개

Model Structure Train part Wavelet filter input Raw signal (normal) output tra in Convolutional Auto Encoder Wavelet filter input output tes t Anomaly Score Raw signal (anomalous) Test part Threshold = θ Score>θ : anomaly (depressed) Score<θ : normal

THANK YOU Q&A