Vector Quantized Time Series Generation with a Bidirectional Prior Model
This addresses the problem of generating realistic time series data for researchers and practitioners, offering a novel approach that overcomes limitations of GANs and RNNs, though it is incremental in applying image generation techniques to time series.
The authors tackled time series generation by introducing TimeVQVAE, the first method using vector quantization techniques, which improved synthetic signal quality with sharper modularity changes compared to existing methods, achieving better scores on metrics like Fréchet inception distance across all UCR archive datasets.
Time series generation (TSG) studies have mainly focused on the use of Generative Adversarial Networks (GANs) combined with recurrent neural network (RNN) variants. However, the fundamental limitations and challenges of training GANs still remain. In addition, the RNN-family typically has difficulties with temporal consistency between distant timesteps. Motivated by the successes in the image generation (IMG) domain, we propose TimeVQVAE, the first work, to our knowledge, that uses vector quantization (VQ) techniques to address the TSG problem. Moreover, the priors of the discrete latent spaces are learned with bidirectional transformer models that can better capture global temporal consistency. We also propose VQ modeling in a time-frequency domain, separated into low-frequency (LF) and high-frequency (HF). This allows us to retain important characteristics of the time series and, in turn, generate new synthetic signals that are of better quality, with sharper changes in modularity, than its competing TSG methods. Our experimental evaluation is conducted on all datasets from the UCR archive, using well-established metrics in the IMG literature, such as Fréchet inception distance and inception scores. Our implementation on GitHub: \url{https://github.com/ML4ITS/TimeVQVAE}.