Semantically-Guided Inference for Conditional Diffusion Models: Enhancing Covariate Consistency in Time Series Forecasting

Rui Ding, Hanyang Meng, Zeyang Zhang, Jielong Yang

Keywords: Time Series Forecasting, Diffusion Models, Inference-Time Correction, Guided Inference

TL;DR: We introduce SemGuide, a plug-and-play inference framework for conditional diffusion models that uses a learned semantic score to guide denoising trajectories via stepwise reweighting, improving covariate consistency without retraining the model.

Abstract: Diffusion models have demonstrated strong performance in time series forecasting, yet often suffer from semantic misalignment between generated trajectories and conditioning covariates, especially under complex or multimodal conditions. To address this issue, we propose SemGuide, a plug-and-play, inference-time method that enhances covariate consistency in conditional diffusion models. Our approach introduces a scoring network to assess the semantic alignment between intermediate diffusion states and future covariates. These scores serve as proxy likelihoods in a stepwise importance reweighting procedure, which progressively adjusts the sampling path without altering the original training process. The method is model-agnostic and compatible with any conditional diffusion framework. Experiments on real-world forecasting tasks show consistent gains in both predictive accuracy and covariate alignment, with especially strong performance under complex conditioning scenarios.

Primary Area: learning on time series and dynamical systems

Submission Number: 7612