IncLSTM: Incremental Ensemble LSTM Model towards Time Series Data. (June 2021)

• Record Type: Journal Article
• Authors: Wang, Huiju; Li, Mengxuan; Yue, Xiao
• Is Part Of: Computers & electrical engineering. Volume 92(2021)
• Publication Date: 2021-06
• Subjects: Incremental learning -- Time series data -- IncLSTM -- FL-Share -- Ensemble learning -- LSTM neural network -- Transfer learning; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107156
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Long short-term memory (LSTM) is one of the most widely used recurrent neural network. Traditionally, it adopts an offline batch mode for model training. To be updated with new data, the network has to be re-trained with merged data using both old and new data, which is very time-consuming and causes catastrophic forgetting. To address this issue, we proposed an incremental ensemble LSTM model-IncLSTM, which fuses ensemble learning and transfer learning to implement incremental updating of the model. The experimental results showed that, in average, the proposed method decreases training time by 18.8%, and improves the prediction accuracy by 15.6% compared with the traditional methods. More importantly, the larger the training data size is, the more efficient IncLSTM would be. While updating the new model, current model predicts independently and concurrently, and the switch between current model and new model occurs once the update is completed, which significantly improves the training efficiency of the model.