发表机构
Nanyang Technological University; Tsinghua University; Shanghai Jiao Tong University（南洋理工大学; 清华大学; 上海交通大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
ProtoFlow提出用训练好的VQ码本构建原型先验，结合流匹配实现非自回归多元时间序列预测，避免AR暴露偏差并加速收敛，在基准上性能优越。
AI 中文摘要
生成式建模在多元时间序列（MTS）预测中展现出强大潜力，尤其是在扩展到高维设置时。基于扩散的方法取得了有竞争力的性能，但通常在推理时需要大量采样步骤。因此，基于VAE的非迭代预测框架作为一种高效替代方案应运而生。在这一研究路线中，向量量化（VQ）通过将多元序列映射为紧凑的离散表示，实现了可控的潜在空间建模。然而，现有的基于VQ的预测方法通常依赖自回归（AR）令牌生成，这存在暴露偏差和训练-推理不匹配的问题。流匹配为潜在预测提供了一种高效的非自回归替代方案，但现有公式通常从通用高斯先验初始化传输。相反，我们观察到训练好的VQ码本已经捕获了具有代表性的潜在原型，因此可以作为流匹配更具信息量的先验。基于这一洞察，我们提出了ProtoFlow，一种结合向量量化自编码与原型先验流匹配的预测框架。我们的方法首先将多元序列映射到离散潜在空间，然后从学习到的码本构建结构化先验，最后学习基于DiT的整流流，将样本从该先验传输到以历史观测为条件的未来潜在表示。通过用学习到的原型先验替换通用噪声初始化，ProtoFlow避免了AR令牌预测的展开不匹配，并促进了更快的训练收敛。在基准数据集上的大量实验表明，它在实现高效推理的同时持续取得了优越的预测性能。
英文摘要
Generative modeling has shown strong promise for multivariate time mseries (MTS) forecasting, especially scale to high-dimensional settings. Diffusion-based methods achieve competitive performance but typically require many sampling steps at inference. VAE-based non-iterative forecasting frameworks have therefore emerged as an efficient alternative. Within this line of work, vector quantization (VQ) enables controllable latent space modeling by mapping multivariate series into compact discrete representations. Existing VQ-based forecasting methods, however, typically rely on autoregressive (AR) token generation, which suffers from exposure bias and training-inference mismatch. Flow matching provides an efficient non-autoregressive alternative for latent forecasting, but existing formulations usually initialize transport from a generic Gaussian prior. We instead observe that the trained VQ codebook already captures representative latent prototypes and can thus serve as a more informative prior for flow matching. Based on this insight, we propose ProtoFlow, a forecasting framework that combines vector-quantized autoencoding with Prototype-prior Flow matching. Our method first maps multivariate sequences into a discrete latent space, then constructs a structured prior from the learned codebook, and finally learns a DiT-based rectified flow to transport samples from this prior to future latent representations conditioned on historical observations. By replacing generic noise initialization with a learned prototype prior, ProtoFlow avoids the rollout mismatch of AR token prediction and promotes faster training convergence. Extensive experiments on benchmark datasets show that it consistently achieves superior forecasting performance with efficient inference.
