TimeBridge: Better Diffusion Prior Design with Bridge Models for Time Series Generation

TL;DR

TimeBridge is proposed, a framework that flexibly synthesizes time series data by using diffusion bridges to learn paths between a chosen prior and the data distribution and experiments show that this framework with data-driven priors outperforms standard diffusion models on time series generation.

Abstract

Time series generation is widely used in real-world applications such as simulation, data augmentation, and hypothesis testing. Recently, diffusion models have emerged as the de facto approach to time series generation, enabling diverse synthesis scenarios. However, the fixed standard-Gaussian diffusion prior may be ill-suited for time series data, which exhibit properties such as temporal order and fixed time points. In this paper, we propose TimeBridge, a framework that flexibly synthesizes time series data by using diffusion bridges to learn paths between a chosen prior and the data distribution. We then explore several prior designs tailored to time series synthesis. Our framework covers (i) data- and time-dependent priors for unconditional generation and (ii) scale-preserving priors for conditional generation. Experiments show that our framework with data-driven priors outperforms standard diffusion models on time series generation.

Figures (6)

• Figure 1: Illustration of the proposed time series diffusion bridge framework enabling various prior selections.
• Figure 2: Performance measures (lower is better) on different tasks and sampling time (lower is better) of time series diffusion models. The proposed TimeBridge achieves both generalization in various tasks and sampling efficiency.
• Figure 3: Illustration of time series generation situations under various conditions.
• Figure 4: Visualization on t-SNE. Red and blue indicate original and synthetic data, respectively.
• Figure 5: Illustration of sampling path from (Left) prior to (Right) data samples on the ETTh.