Intra-domain and cross-domain transfer learning for time series data—How transferable are the features?

作者：

Highlights：

• Transfer learning is very likely to either result in positive or non-negative effects.

• The hyperparameters for optimal transfer learning highly depend on the chosen model.

• A model pretrained on an unrelated task can be better than a randomly initialised model.

摘要

•Transfer learning is very likely to either result in positive or non-negative effects.•The hyperparameters for optimal transfer learning highly depend on the chosen model.•A model pretrained on an unrelated task can be better than a randomly initialised model.

论文关键词：Machine learning,Transfer learning,Time series,Fine-tuning,Convolutional neural networks

论文评审过程：Received 22 August 2021, Revised 7 November 2021, Accepted 13 December 2021, Available online 24 December 2021, Version of Record 12 January 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.107976