Loading...

Exploiting Transfer Learning in Deep Neural Networks for Time Series

Salami, Mohammad Sadegh | 2020

810 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53156 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Manzuri Shalmani, Mohammad Taghi
  7. Abstract:
  8. The importance of transfer learning in image-related problems comes from its many advantages that are sometimes undeniable. Previous researches have well shown the success of transfer learning in this area using deep neural networks. However, transfer learning for time series data has not yet been done in a conventional and automated manner. The main reason for avoiding transfer learning in this domain relates to the dynamic and stochastic nature of the time series, where they show a time-varying behavior. Previous experiments have shown that transfer learning between two heterogeneous time series could harm the forecasting accuracy of a model. Therefore, in this thesis, we aim to explore and develop a set of principles for transfer learning in time series data, and in particular, we have explored possible methods to accurately extract useful transferable features by learning a common representation among different time series. Then we have compared and analyzed the effect of transfer learning in these methods by transferring the obtained knowledge to improve the forecasting performance of the model in other time series. The experiments show the effectiveness of the proposed methods in exploiting transfer learning for time series: Our results show an average improvement of 20% compared to the results of the existing methods
  9. Keywords:
  10. Representation Learning ; Time Series ; Forecasting ; Deep Neural Networks ; Transfer Learning ; Feature Extraction

 Digital Object List

 Bookmark