Multivariate time series forecasting is a fundamental problem in sequence modeling, with broad applications spanning energy systems, meteorology, finance, and transportation. Despite significant progress in Transformer-based, TCN-based, and MLP-based architectures, three persistent neural network design challenges remain unresolved: 1) non-adaptive frequency-domain filtering that applies identical spectral parameters across heterogeneous variables; 2) inadequate utilization of informative short-term fluctuations that are discarded as noise yet encode critical event-driven dynamics; and 3) insufficient exploration of inter-scale dependencies in multi-scale temporal representations. To address these challenges, this paper proposes AGformer, a novel forecasting architecture integrating three complementary neural modules: 1) A variable-oriented Adaptive Multi-k Filter (AKF) that dynamically generates per-variable spectral filtering parameters from learned frequency-domain features, eliminating the need for manual global hyperparameter tuning. 2) A Dual-Gate Fusion-Attention (DGFA) mechanism that synergistically models long-term trends and short-term fluctuations via complementary temporal and content gating, recovering event-driven signals that are otherwise discarded by standard denoising. 3) A Multi-Scale Embedding (MSED) layer that fuses temporal patterns across resolutions through attention-based inter-scale weighting, capturing cross-resolution hierarchical dependencies that simple concatenation or summation cannot model. Extensive experiments on nine representative benchmark datasets across electricity, weather, finance, and transportation domains demonstrate AGformer's consistent superiority, achieving average reductions of 8.1%-23.6% in MSE and MAE relative to state-of-the-art baselines across nine benchmarks.
Combining electronic health records (EHR) with medical images is important for improving the diagnosis of Alzheimer's disease (AD). However, neuroimages and EHR differ greatly in semantic content and data distribution. Neuroimages mainly reflect brain structure and metabolic abnormalities, while EHR records clinical symptoms and behavioral changes. Directly fusing these heterogeneous modalities may mix shared disease information with modality-specific diagnostic cues, resulting in redundant and ambiguous representations. To address this problem, we propose DA-S2RL, a disease-aware shared-specific multimodal representation learning framework for AD diagnosis. DA-S2RL organizes heterogeneous image and EHR features into disease-related shared semantics and modality-specific diagnostic residuals. The core of DA-S2RL is the shared-specific cross-modal coding (SCC) module. It strengthens cross-modal semantic consistency, suppresses redundant correlations, and enables mutual refinement between shared semantics and diagnostic residuals. Before SCC, image features are integrated into a compact image representation, and EHR features are encoded into the same feature space. Experiments on ADNI and NACC demonstrate that explicitly organizing heterogeneous image-EHR representations into shared disease semantics and modality-specific residuals improves multimodal diagnostic learning under different image-EHR settings.
