| Autor: |
Koparanov, Kiril, Trifonov, Rumen, Minkovska, Daniela, Georgiev, Krasin |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2022, Vol. 2557 Issue 1, p1-9, 9p |
| Abstrakt: |
Predicting long sequences of values is a task with multiple applications. Aircraft manufacturers and airlines stock prices reflect the society and investors trust in aviation industry. The amount of transported freight is directly related to logistics planning. Exploiting neural networks for such a task has not been studied in depth. In some areas, a recurrent approach is used, e.g. for text generation, but other studies show that direct output of the entire sequence in a single step is more accurate. A comprehensive theoretical analysis of the properties of such models is currently beyond the capabilities of modern science, so we will limit ourselves to an empirical study of specific cases with synthetic data and real financial multivariate time series. The performance of the direct and recurrent approaches is compared and conclusions are made about their effectiveness under different conditions. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
