Time Series Super Resolution withTemporal Adaptive Batch Normalization

Volodymyr Kuleshov, Sawyer Birnbaum, Zayd Enam, Pang Wei Koh, Stefano Ermon

Abstract: In many machine learning problems, data is naturally expressed as a time series. Here, we introduce a deep neural network architecture for reconstructing a high-resolution time series signal from low-resolution measurements, a task that we call time series super resolution. Central to our architecture is a novel temporal adaptive normalization layer that combines the strength of convolutional and recurrent approaches. We apply our model to diverse super resolution problems: audio super-resolution and the enhancement of functional genomics assays. In each case, our method significantly outperforms strong baselines, demonstrating its ability to solve practical problems in a wide range of domains.