TimeMIL: Advancing Multivariate Time Series Classification via a Time-aware Multiple Instance Learning

• DOI: 10.48550/arxiv.2405.03140
• Publication Date: 2024-05-05
• Authors: Xiwen Chen; Peijie Qiu; Wenhui Zhu; Huayu Li; Hao Wang; Aristeidis Sotiras; Yalin Wang; Abolfazl Razi
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Machine Learning (cs.LG)

Deep neural networks, including transformers and convolutional have significantly improved multivariate time series classification (MTSC). However, these methods often rely on supervised learning, which does not fully account for the sparsity locality of patterns in data (e.g., diseases-related anomalous points ECG). To address this challenge, we formally reformulate MTSC as a weakly problem, introducing novel multiple-instance learning (MIL) framework better localization interest modeling dependencies within series. Our approach, TimeMIL, formulates temporal correlation ordering time-aware MIL pooling, leveraging tokenized transformer with specialized learnable wavelet positional token. The proposed method surpassed 26 recent state-of-the-art methods, underscoring effectiveness TimeMIL MTSC.