Sequence prediction using recurrent neural network
The project implemented a Gap-Filling Engine capable of filling in gaps in missing sequences of various types. A strategy was introduced to look forward into subsequent data, enabling the Engine to improve the accuracy of the prediction by more than 30%. A Sequence Model based on Long Short-term Mem...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Final Year Project |
| 語言: | English |
| 出版: |
2017
|
| 主題: | |
| 在線閱讀: | http://hdl.handle.net/10356/70503 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Nanyang Technological University |
| 語言: | English |
| 總結: | The project implemented a Gap-Filling Engine capable of filling in gaps in missing sequences of various types. A strategy was introduced to look forward into subsequent data, enabling the Engine to improve the accuracy of the prediction by more than 30%. A Sequence Model based on Long Short-term Memory (LSTM) Recurrent Neural Network was used to learn patterns in several sequence types.
Optimization technique by caching the LSTM was implemented, shortening runtime by several hundred times, allowing significantly more experiments to be executed. Performance evaluation strategy was designed and carried out to analyze the impact of various factors on the Gap-Filling Engine, providing a better understand of the Engine, as well as recommendations for users. The Gap-Filling Engine shows potentials of applications on several fields, especially in reconstructing structured sequential data with missing values. |
|---|