arXiv:2512.07873v1 Announce Type: new 
Abstract: Recent studies show that using diffusion models for time series signal reconstruc- tion holds great promise. However, such approaches remain largely unexplored in the domain of medical time series. The unique characteristics of the physiological time series signals, such as multivariate, high temporal variability, highly noisy, and artifact-prone, make deep learning-based approaches still challenging for tasks such as imputation. Hence, we propose a novel Mixture of Experts (MoE)-based noise estimator within a score-based diffusion framework. Specifically, the Receptive Field Adaptive MoE (RFAMoE) module is designed to enable each channel to adap- tively select desired receptive fields throughout the diffusion process. Moreover, recent literature has found that when generating a physiological signal, performing multiple inferences and averaging the reconstructed signals can effectively reduce reconstruction errors, but at the cost of significant computational and latency over- head. We design a Fusion MoE module and innovatively leverage the nature of MoE module to generate K noise signals in parallel, fuse them using a routing mechanism, and complete signal reconstruction in a single inference step. This design not only improves performance over previous methods but also eliminates the substantial computational cost and latency associated with multiple inference processes. Extensive results demonstrate that our proposed framework consistently outperforms diffusion-based SOTA works on different tasks and datasets.

تم تسليط الضوء على التقدم الأخير في إعادة بناء إشارات السلاسل الزمنية من خلال تقديم مُقدّر ضوضاء جديد يعتمد على مزيج من الخبراء (MoE) ضمن إطار عمل قائم على الانتشار، يستهدف بشكل خاص التحديات المتعلقة بإشارات السلاسل الزمنية الفسيولوجية. يهدف هذا النهج إلى تحسين دقة مهام الإحلال، التي تعتبر حاسمة في التطبيقات الطبية حيث يمكن أن تكون البيانات متعددة المتغيرات وصاخبة.

Se han destacado avances recientes en la reconstrucción de señales de series temporales mediante la introducción de un nuevo estimador de ruido basado en un Mezcla de Expertos (MoE) dentro de un marco de difusión basado en puntajes, dirigido específicamente a los desafíos de las señales de series temporales fisiológicas. Este enfoque busca mejorar la precisión de las tareas de imputación, que son críticas en aplicaciones médicas donde los datos pueden ser multivariados y ruidosos.

Des avancées récentes dans la reconstruction des signaux de séries temporelles ont été mises en avant avec l'introduction d'un nouvel estimateur de bruit basé sur un mélange d'experts (MoE) dans un cadre de diffusion basé sur des scores, ciblant spécifiquement les défis des signaux de séries temporelles physiologiques. Cette approche vise à améliorer la précision des tâches d'imputation, qui sont critiques dans les applications médicales où les données peuvent être multivariées et bruyantes.

Recent advancements in time series signal reconstruction have been highlighted through the introduction of a novel Mixture of Experts (MoE)-based noise estimator within a score-based diffusion framework, specifically targeting the challenges of physiological time series signals. This approach aims to enhance the accuracy of imputation tasks, which are critical in medical applications where data can be multivariate and noisy.

Advancing physiological time series reconstruction and imputation via mixture of receptive fields and experts fusion

Was this article worth reading? Share it

LucidQuery AI

Magicley AI

Augmeta