arXiv:2512.14718v1 Announce Type: new 
Abstract: Effective multivariate time series forecasting often benefits from accurately modeling complex inter-variable dependencies. However, existing attention- or graph-based methods face three key issues: (a) strong temporal self-dependencies are often disrupted by irrelevant variables; (b) softmax normalization ignores and reverses negative correlations; (c) variables struggle to perceive their temporal positions. To address these, we propose \textbf{SEED}, a Spectral Entropy-guided Evaluation framework for spatial-temporal Dependency modeling. SEED introduces a Dependency Evaluator, a key innovation that leverages spectral entropy to dynamically provide a preliminary evaluation of the spatial and temporal dependencies of each variable, enabling the model to adaptively balance Channel Independence (CI) and Channel Dependence (CD) strategies. To account for temporal regularities originating from the influence of other variables rather than intrinsic dynamics, we propose Spectral Entropy-based Fuser to further refine the evaluated dependency weights, effectively separating this part. Moreover, to preserve negative correlations, we introduce a Signed Graph Constructor that enables signed edge weights, overcoming the limitations of softmax. Finally, to help variables perceive their temporal positions and thereby construct more comprehensive spatial features, we introduce the Context Spatial Extractor, which leverages local contextual windows to extract spatial features. Extensive experiments on 12 real-world datasets from various application domains demonstrate that SEED achieves state-of-the-art performance, validating its effectiveness and generality.

تم اقتراح إطار جديد يسمى SEED للتنبؤ بالسلاسل الزمنية متعددة المتغيرات، حيث يعالج القضايا الرئيسية في الأساليب الحالية المتعلقة بالاعتماد الذاتي الزمني، وتطبيع الارتباطات، وإدراك المتغيرات لمواقعها الزمنية. يقدم SEED مُقيِّم الاعتماد الذي يستخدم الانتروبيا الطيفية لتقييم الاعتمادات المكانية والزمنية بشكل ديناميكي.

Se ha propuesto un nuevo marco llamado SEED para la predicción de series temporales multivariadas, abordando problemas clave en los métodos existentes relacionados con las auto-dependencias temporales, la normalización de correlaciones y la percepción de las posiciones temporales por parte de las variables. SEED introduce un Evaluador de Dependencia que utiliza la entropía espectral para evaluar dinámicamente las dependencias espaciales y temporales.

Un nouveau cadre nommé SEED a été proposé pour la prévision de séries temporelles multivariées, abordant des problèmes clés dans les méthodes existantes liés aux auto-dépendances temporelles, à la normalisation des corrélations et à la perception des positions temporelles par les variables. SEED introduit un Évaluateur de Dépendance qui utilise l'entropie spectrale pour évaluer dynamiquement les dépendances spatiales et temporelles.

A new framework named SEED has been proposed for multivariate time series forecasting, addressing key issues in existing methods related to temporal self-dependencies, normalization of correlations, and variable perception of temporal positions. SEED introduces a Dependency Evaluator that utilizes spectral entropy to assess spatial and temporal dependencies dynamically.

SEED: Spectral Entropy-Guided Evaluation of SpatialTemporal Dependencies for Multivariate Time Series Forecasting

arXiv:2601.07858v1 Announce Type: new 
Abstract: Generalisation to unseen subjects in EEG-based emotion classification remains a challenge due to high inter-and intra-subject variability. Continual learning (CL) poses a promising solution by learning from a sequence of tasks while mitigating catastrophic forgetting. Regularisation-based CL approaches, such as Elastic Weight Consolidation (EWC), Synaptic Intelligence (SI), and Memory Aware Synapses (MAS), are commonly used as baselines in EEG-based CL studies, yet their suitability for this problem remains underexplored. This study theoretically and empirically finds that regularisation-based CL methods show limited performance for EEG-based emotion classification on the DREAMER and SEED datasets. We identify a fundamental misalignment in the stability-plasticity trade-off, where regularisation-based methods prioritise mitigating catastrophic forgetting (backward transfer) over adapting to new subjects (forward transfer). We investigate this limitation under subject-incremental sequences and observe that: (1) the heuristics for estimating parameter importance become less reliable under noisy data and covariate shift, (2) gradients on parameters deemed important by these heuristics often interfere with gradient updates required for new subjects, moving optimisation away from the minimum, (3) importance values accumulated across tasks over-constrain the model, and (4) performance is sensitive to subject order. Forward transfer showed no statistically significant improvement over sequential fine-tuning (p > 0.05 across approaches and datasets). The high variability of EEG signals means past subjects provide limited value to future subjects. Regularisation-based continual learning approaches are therefore limited for robust generalisation to unseen subjects in EEG-based emotion classification.

تشير دراسة حديثة إلى قيود أساليب التعلم المستمر المعتمدة على التنظيم في تصنيف المشاعر باستخدام تخطيط الدماغ الكهربائي (EEG)، خاصة في تعميم النتائج على مواضيع غير معروفة. توضح الأبحاث أن الأساليب مثل دمج الأوزان المرنة والذكاء المشبكي تعطي الأولوية لتقليل النسيان الكارثي على التكيف مع مواضيع جديدة، مما يؤدي إلى أداء دون المستوى الأمثل على مجموعات بيانات DREAMER وSEED.

Un estudio reciente destaca las limitaciones de los métodos de aprendizaje continuo basados en regularización en la clasificación de emociones mediante EEG, especialmente en la generalización a sujetos no vistos. La investigación indica que enfoques como la Consolidación de Pesos Elásticos y la Inteligencia Sináptica priorizan la mitigación del olvido catastrófico sobre la adaptación a nuevos sujetos, resultando en un rendimiento subóptimo en los conjuntos de datos DREAMER y SEED.

Une étude récente met en lumière les limitations des méthodes d'apprentissage continu basées sur la régularisation dans la classification des émotions par EEG, en particulier en ce qui concerne la généralisation à des sujets non vus. La recherche indique que des approches comme la consolidation des poids élastiques et l'intelligence synaptique privilégient la réduction de l'oubli catastrophique au détriment de l'adaptation à de nouveaux sujets, entraînant des performances sous-optimales sur les ensembles de données DREAMER et SEED.

A recent study highlights the limitations of regularisation-based continual learning methods in EEG-based emotion classification, particularly in generalising to unseen subjects. The research indicates that approaches like Elastic Weight Consolidation and Synaptic Intelligence prioritize mitigating catastrophic forgetting over adapting to new subjects, resulting in suboptimal performance on the DREAMER and SEED datasets.

SEED: Spectral Entropy-Guided Evaluation of SpatialTemporal Dependencies for Multivariate Time Series Forecasting

Was this article worth reading? Share it

LucidQuery AI

Graphite Note

The Visualizer

Attentive AI

Supametas.AI

Syft AI

Ready to build your own newsroom?