Multiscale echo self-attention memory network for multivariate time series classification

Recently, ESN has been applied to time series classification own to its high-dimensional random projec-tion ability and training efficiency characteristic. The major drawback of applying ESN to time series clas-sification is that ESN cannot capture long-term dependency information well. Therefore, the Multiscale Echo Self-Attention Memory Network (MESAMN) is proposed to address this issue. Specifically, the MESAMN consists of a memory encoder and a memory learner. In the memory encoder, multiple differ-ently initialized ESNs are utilized for high-dimensional projection which is then followed by a self -attention mechanism to capture the long-term dependent features. A multiscale convolutional neural network is developed as the memory learner to learn local features using features extracted by the mem-ory encoder. Experimental results show that the proposed MESAMN yields better performance on 18 multivariate time series classification tasks as well as three 3D skeleton-based action recognition tasks compared to existing models. Furthermore, the capacity for capturing long-term dependencies of the MESAMN is verified empirically.(c) 2022 Elsevier B.V. All rights reserved.

Automated summary

Recently, ESN has been applied to time series classification for its high-dimensional random projection ability and training efficiency characteristic. However, the major drawback of ESN is its inability to capture long-term dependency information well. To address this issue, the Multiscale Echo Self-Attention Memory Network (MESAMN) is proposed, which consists of a memory encoder and a memory learner. The experimental results show that MESAMN outperforms existing models in various time series classification tasks and 3D skeleton-based action recognition tasks, and its capacity for capturing long-term dependencies is empirically verified.