Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TimeVAE: A Variational Auto-Encoder for Multivariate Time Series Generation

Abhyuday Desai, Cynthia Freeman, Zuhui Wang, Ian Beaver

TL;DR

TimeVAE introduces a variational auto-encoder tailored for multivariate time series, offering a base generative model and an interpretable variant with explicit trend and seasonality components. The approach achieves high fidelity to real data while delivering competitive next-step predictions and significantly reduced training times compared to GAN-based methods. By enabling domain-informed temporal constructs, TimeVAE provides interpretable outputs and denoising benefits that improve downstream tasks. Across four datasets and varying data availability, TimeVAE demonstrates robust performance, particularly under data-scarce regimes, highlighting its practical value for privacy-aware and scenario-driven synthetic data generation.

Abstract

Recent work in synthetic data generation in the time-series domain has focused on the use of Generative Adversarial Networks. We propose a novel architecture for synthetically generating time-series data with the use of Variational Auto-Encoders (VAEs). The proposed architecture has several distinct properties: interpretability, ability to encode domain knowledge, and reduced training times. We evaluate data generation quality by similarity and predictability against four multivariate datasets. We experiment with varying sizes of training data to measure the impact of data availability on generation quality for our VAE method as well as several state-of-the-art data generation methods. Our results on similarity tests show that the VAE approach is able to accurately represent the temporal attributes of the original data. On next-step prediction tasks using generated data, the proposed VAE architecture consistently meets or exceeds performance of state-of-the-art data generation methods. While noise reduction may cause the generated data to deviate from original data, we demonstrate the resulting de-noised data can significantly improve performance for next-step prediction using generated data. Finally, the proposed architecture can incorporate domain-specific time-patterns such as polynomial trends and seasonalities to provide interpretable outputs. Such interpretability can be highly advantageous in applications requiring transparency of model outputs or where users desire to inject prior knowledge of time-series patterns into the generative model.

TimeVAE: A Variational Auto-Encoder for Multivariate Time Series Generation

TL;DR

TimeVAE introduces a variational auto-encoder tailored for multivariate time series, offering a base generative model and an interpretable variant with explicit trend and seasonality components. The approach achieves high fidelity to real data while delivering competitive next-step predictions and significantly reduced training times compared to GAN-based methods. By enabling domain-informed temporal constructs, TimeVAE provides interpretable outputs and denoising benefits that improve downstream tasks. Across four datasets and varying data availability, TimeVAE demonstrates robust performance, particularly under data-scarce regimes, highlighting its practical value for privacy-aware and scenario-driven synthetic data generation.

Abstract

Recent work in synthetic data generation in the time-series domain has focused on the use of Generative Adversarial Networks. We propose a novel architecture for synthetically generating time-series data with the use of Variational Auto-Encoders (VAEs). The proposed architecture has several distinct properties: interpretability, ability to encode domain knowledge, and reduced training times. We evaluate data generation quality by similarity and predictability against four multivariate datasets. We experiment with varying sizes of training data to measure the impact of data availability on generation quality for our VAE method as well as several state-of-the-art data generation methods. Our results on similarity tests show that the VAE approach is able to accurately represent the temporal attributes of the original data. On next-step prediction tasks using generated data, the proposed VAE architecture consistently meets or exceeds performance of state-of-the-art data generation methods. While noise reduction may cause the generated data to deviate from original data, we demonstrate the resulting de-noised data can significantly improve performance for next-step prediction using generated data. Finally, the proposed architecture can incorporate domain-specific time-patterns such as polynomial trends and seasonalities to provide interpretable outputs. Such interpretability can be highly advantageous in applications requiring transparency of model outputs or where users desire to inject prior knowledge of time-series patterns into the generative model.
Paper Structure (15 sections, 4 figures, 3 tables)

This paper contains 15 sections, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methods
  4. Variational Auto-Encoder as a Generative Model
  5. VAE Loss Function
  6. Base TimeVAE Architecture
  7. Interpretable TimeVAE
  8. TimeVAE Objective Function
  9. Experiments
  10. Data Sets
  11. Comparison Methodologies
  12. Comparison Metrics
  13. Procedure
  14. Results/Discussion
  15. Conclusions

Figures (4)

  • Figure 1: Block diagram of components in Base TimeVAE
  • Figure 2: Block diagram of components in Interpretable TimeVAE
  • Figure 3: Trend and Seasonality Blocks in Interpretable TimeVAE
  • Figure 4: t-SNE plots for the TimeGAN (top left), TimeVAE (top right), RCGAN (bottom left), and T-Forcing (bottom right) models for the 4 data sets under various training size percentages ($p$). Red is for original data, and blue for synthetic data. An empty t-SNE chart appears when not enough training data is present for the model to generate synthetic data.