ProjectsProject Details

The goal of this work is to use ordinary differential equation (ODE) network for sleeping stage identification of the incoming measurement. Neural ODEs are neural network models which generalize standard layer to layer propagation to continuous depth models. In our work, the data is a time series of EEG measurements during people sleeping time. We want to feed this data to the model and see the performance of our model in classifying and predicting sleeping stages.

To achieve this goal, we implemented a neural network architecture which includes the Neural ODE model. Then the model was trained for the classification of the sleeping data to different sleeping stages and tested on an unseen data. Few feature engineering methods were applied on the data before training the model. One method was to pass the original channels through band pass filters according to the wave pattern assumed to be in sleeping EEG signals. To analyze the outcome of this method, 2 ablation studies were conducted by removing few input channels. Other feature engineering method was to pass to the model a frequency domain signals with Fast Fourier transform (FFT) and Power spectral density (PSD) computed on the original time domain signals. The experiments were done per subject and inter-subject wise. Overall performance of the model was good, and ODE network capability to handle EEG signals was proven. Indeed, ODE network succeeded in learning the dynamics of the input. The ablation study showed the significance of the delta frequency band for the classification task of the model.