arXiv:2512.15116v2 Announce Type: replace Abstract: Multivariate time series imputation is fundamental in applications such as healthcare, traffic forecasting, and biological modeling, where sensor failures and irregular sampling lead to pervasive missing values. However, existing Transformer- and diffusion-based models lack explicit inductive biases and frequency awareness, limiting their generalization under structured missing patterns and distribution shifts. We propose FADTI, a diffusion-based framework that injects frequency-informed feature modulation via a learnable Fourier Bias Project
The proliferation of sensors and the increased reliance on data-driven decision-making across various industries make robust multivariate time series imputation essential, especially with growing data complexities and missing values.
Improved imputation techniques for multivariate time series are critical for advancing AI applications in sectors like healthcare, finance, and logistics, enhancing the reliability and accuracy of predictive models.
This research introduces a novel diffusion-based model with explicit frequency awareness, potentially leading to more accurate and generalizable AI models, particularly in scenarios with structured missing data and distribution shifts.
- · AI/ML researchers
- · Healthcare providers
- · Logistics and supply chain companies
- · Financial institutions
- · Organizations relying on less sophisticated imputation methods
- · Sectors with high data loss rates and poor AI model performance
More accurate and resilient multivariate time series analyses become broadly available.
Enhanced data quality improves the performance and trustworthiness of AI systems in critical applications, accelerating their adoption.
Industries reliant on high-fidelity time series data experience significant improvements in efficiency, forecasting accuracy, and decision-making capabilities, leading to new service offerings and operational benchmarks.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG