Correlative Channel-Aware Fusion for Multi-View Time Series Classification

Multi-view time series classification (MVTSC) aims to improve the performance by fusing distinctive temporal information from multiple views. Existing methods for MVTSC mainly aim fuse multi-view at an early stage, e.g., extracting a common feature subspace among However, these approaches may not fully explore unique patterns of each view in complicated series. Additionally, label correlations views, which are critical boosting, usually under-explored problem. To address aforementioned issues, we propose Correlative Channel-Aware Fusion (C$^2$AF) network. First, C$^2$AF extracts comprehensive and robust two-stream structured encoder view, derives intra-view/inter-view with concise correlation matrix. Second, channel-aware learnable fusion mechanism is implemented through CNN further global correlative patterns. Our end-to-end framework MVTSC. Extensive experimental results on three real-world datasets demonstrate superiority our over state-of-the-art methods. A detailed ablation study also provided illustrate indispensability model component.