Bidirectional Recurrent Neural Networks as Generative Models - Reconstructing Gaps in Time Series
This work addresses the challenge of efficient gap-filling in time series for domains like text and music, offering a novel unsupervised approach that is more scalable than complex Bayesian inference.
The paper tackled the problem of reconstructing missing gaps in high-dimensional time series with complex dynamics, proposing two probabilistic interpretations of bidirectional RNNs that significantly improved accuracy over unidirectional methods, achieving much higher accuracy on text data and scalable performance on music data.
Bidirectional recurrent neural networks (RNN) are trained to predict both in the positive and negative time directions simultaneously. They have not been used commonly in unsupervised tasks, because a probabilistic interpretation of the model has been difficult. Recently, two different frameworks, GSN and NADE, provide a connection between reconstruction and probabilistic modeling, which makes the interpretation possible. As far as we know, neither GSN or NADE have been studied in the context of time series before. As an example of an unsupervised task, we study the problem of filling in gaps in high-dimensional time series with complex dynamics. Although unidirectional RNNs have recently been trained successfully to model such time series, inference in the negative time direction is non-trivial. We propose two probabilistic interpretations of bidirectional RNNs that can be used to reconstruct missing gaps efficiently. Our experiments on text data show that both proposed methods are much more accurate than unidirectional reconstructions, although a bit less accurate than a computationally complex bidirectional Bayesian inference on the unidirectional RNN. We also provide results on music data for which the Bayesian inference is computationally infeasible, demonstrating the scalability of the proposed methods.