Entropy of LOO Predictable Horizons to Select a Learning Machine and a Representative Prediction of Chaotic Time Series
In: Communications in Computer and Information Science ISBN: 9783030638221 ICONIP (5); (2020)
Online
unknown
Zugriff:
Recently, we have presented several methods to select representative predictions of chaotic time series. Here, the methods employ strong learners capable of making predictions with small error, where usual ensemble mean does not work well owing to an ensemble member with short term predictable horizon because the ensemble prediction error of chaotic time series grows exponentially after the smallest predictable horizon of the ensemble member. Here, we refer to ‘predictable horizon’ as the first point in time after which the prediction error exceeds a certain error threshold. So far, we have developed several methods to select representative predictions from a set of many predictions by means of using a LOOCV (leave-one-out cross-validation) measure to estimate predictable horizon. From the analysis of the methods showing that the method works well with sufficiently large number of predictions generated by sufficiently strong learning machines, this paper presents a method to generate a large number of predictions and select a learning machine and a representative prediction using the entropy of LOO predictable horizons. By means of numerical experiments, we show and examine the effectiveness of the present method.
Titel: |
Entropy of LOO Predictable Horizons to Select a Learning Machine and a Representative Prediction of Chaotic Time Series
|
---|---|
Autor/in / Beteiligte Person: | Miyazaki, Daichi ; Matsuo, Kazuya ; Kurogi, Shuichi |
Link: | |
Quelle: | Communications in Computer and Information Science ISBN: 9783030638221 ICONIP (5); (2020) |
Veröffentlichung: | Springer International Publishing, 2020 |
Medientyp: | unknown |
ISBN: | 978-3-030-63822-1 (print) |
DOI: | 10.1007/978-3-030-63823-8_88 |
Schlagwort: |
|
Sonstiges: |
|