Fig. 3From: Representation learning for clinical time series prediction tasks in electronic health recordsThe architecture of our proposed RNNDAE model. Multi-hot vectors (xt) with time series are added by a Gaussian noise and then encoded by a GRUencoder model into the patient vector (c). Given the patient vector, another GRUdecoder model is used to decode in order to make the input (xt) and the output (yt) are consistent as much as possibleBack to article page