Variational Recurrent Auto-Encoders
This addresses the problem of efficient, large-scale unsupervised learning for time series data, though it appears incremental as it combines existing techniques.
The paper tackles unsupervised learning on time series data by proposing the Variational Recurrent Auto-Encoder (VRAE), which combines RNNs and SGVB to map data to latent representations and generate new data, with a key contribution of using unlabeled data to initialize weights for supervised RNN training.
In this paper we propose a model that combines the strengths of RNNs and SGVB: the Variational Recurrent Auto-Encoder (VRAE). Such a model can be used for efficient, large scale unsupervised learning on time series data, mapping the time series data to a latent vector representation. The model is generative, such that data can be generated from samples of the latent space. An important contribution of this work is that the model can make use of unlabeled data in order to facilitate supervised training of RNNs by initialising the weights and network state.