Less Is More: Generating Time Series with LLaMA-Style Autoregression in Simple Factorized Latent Spaces
This addresses the need for efficient and flexible generative models for time series data, useful in applications like data augmentation and simulation, though it appears incremental as it builds on existing factorization and Transformer methods.
The paper tackled the problem of slow and fixed-length generation in multivariate time series models by proposing FAR-TS, a framework that uses factorized latent spaces and autoregressive Transformers, achieving orders-of-magnitude faster generation than Diffusion-TS while maintaining quality.
Generative models for multivariate time series are essential for data augmentation, simulation, and privacy preservation, yet current state-of-the-art diffusion-based approaches are slow and limited to fixed-length windows. We propose FAR-TS, a simple yet effective framework that combines disentangled factorization with an autoregressive Transformer over a discrete, quantized latent space to generate time series. Each time series is decomposed into a data-adaptive basis that captures static cross-channel correlations and temporal coefficients that are vector-quantized into discrete tokens. A LLaMA-style autoregressive Transformer then models these token sequences, enabling fast and controllable generation of sequences with arbitrary length. Owing to its streamlined design, FAR-TS achieves orders-of-magnitude faster generation than Diffusion-TS while preserving cross-channel correlations and an interpretable latent space, enabling high-quality and flexible time series synthesis.