SDFlow: Similarity-Driven Flow Matching for Time Series Generation

2026-05-07

AI Summary

This work addresses the error accumulation in autoregressive vector quantized models caused by exposure bias during long-sequence generation. To mitigate this issue, the authors propose a non-autoregressive generation framework that leverages flow matching within a frozen VQ latent space, replacing step-by-step prediction with a global transport mapping for parallel sequence synthesis. The approach introduces a novel similarity-driven flow matching mechanism, integrating low-rank manifold decomposition, learnable anchor-point priors, and variational flow matching with discrete supervision to effectively model high-dimensional discrete spaces. Experimental results demonstrate that the proposed method significantly outperforms existing models on long-sequence generation tasks, achieving higher discriminative scores, lower Context-FID, and substantially faster inference speeds compared to autoregressive baselines.

Abstract

Vector quantization (VQ) with autoregressive (AR) token modeling is a widely adopted and highly competitive paradigm for time-series generation. However, such models are fundamentally limited by exposure bias: during inference, errors can accumulate across sequential predictions, leading to pronounced quality degradation in long-horizon generation. To address this, we propose SDFlow ($\textbf{S}$imilarity-$\textbf{D}$riven $\textbf{Flow}$ Matching), a non-autoregressive framework that operates entirely in the frozen VQ latent space and enables parallel sequence generation via flow matching. We tackle three key challenges in making this transition: (1) eliminating exposure bias by replacing step-wise token prediction with a global transport map; (2) mitigating the high-dimensionality of VQ token spaces via a low-rank manifold decomposition with a learned anchor prior over the latent manifold; and (3) incorporating discrete supervision into continuous transport dynamics by introducing a categorical posterior over codebook indices within a variational flow-matching formulation. Extensive experiments show that SDFlow achieves state-of-the-art performance, improving Discriminative Score and substantially reducing Context-FID, particularly for challenging long-sequence generation. Moreover, SDFlow provides significant inference speedups over autoregressive baselines, offering both high fidelity and computational efficiency. Code is available at https://anonymous.4open.science/r/SDFlow-D6F3/

Problem

• exposure bias
• time series generation
• vector quantization
• autoregressive modeling
• long-horizon generation

Innovation

• flow matching
• vector quantization
• non-autoregressive generation
• exposure bias
• latent manifold