WaveStitch: Flexible and Fast Conditional Time Series Generation with Diffusion Models
This work solves the problem of flexible and fast conditional time series generation for forecasting, imputation, and generative tasks, with incremental improvements in combining existing techniques.
The paper tackled the problem of generating conditional time series data by addressing limitations in existing methods, such as handling both metadata and observed signals, and balancing speed and coherence, resulting in WaveStitch achieving 1.81x lower mean-squared-error than state-of-the-art and generating data up to 166.48x faster than autoregressive methods.
Generating temporal data under conditions is crucial for forecasting, imputation, and generative tasks. Such data often has metadata and partially observed signals that jointly influence the generated values. However, existing methods face three key limitations: (1) they condition on either the metadata or observed values, but rarely both together; (2) they adopt either training-time approaches that fail to generalize to unseen scenarios, or inference-time approaches that ignore metadata; and (3) they suffer from trade-offs between generation speed and temporal coherence across time windows--choosing either slow but coherent autoregressive methods or fast but incoherent parallel ones. We propose WaveStitch, a novel diffusion-based method to overcome these hurdles through: (1) dual-sourced conditioning on both metadata and partially observed signals; (2) a hybrid training-inference architecture, incorporating metadata during training and observations at inference via gradient-based guidance; and (3) a novel pipeline-style paradigm that generates time windows in parallel while preserving coherence through an inference-time conditional loss and a stitching mechanism. Across diverse datasets, WaveStitch demonstrates adaptability to arbitrary patterns of observed signals, achieving 1.81x lower mean-squared-error compared to the state-of-the-art, and generates data up to 166.48x faster than autoregressive methods while maintaining coherence. Our code is available at: https://github.com/adis98/WaveStitch