Neural networks are becoming more and more popular for the analysis of time-series. The most successful deep learning systems in this combine convolutional and recurrent layers to extract useful features to temporal relations. Unfortunately, these recurrent models are difficult tune and optimize. In our experience, they often require task-specific, which makes them challenging to use for non-experts. We propose-Time, a fully feed-forward deep learning approach to physiological time segmentation developed for the analysis of sleep data. U-Time is a fully convolutional network based on the U-Net architecture that was proposed for image segmentation. U-Time maps sequential inputs of length to sequences of class labels on a freely chosen temporal. This is done by implicitly classifying every individual time-point of input signal and aggregating these classifications over fixed intervals to the final predictions. We evaluated U-Time for sleep stage classification a large collection of sleep electroencephalography (EEG) datasets. In all, we found that U-Time reaches or outperforms current state-of-the-art learning models while being much more robust in the training process and requiring architecture or hyperparameter adaptation across tasks.
No takes yet. Share an insight, caveat, or question.
Perslev et al. (2019) studied this question.