Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | 2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Proceedings |
| Herausgeber (Verlag) | Institute of Electrical and Electronics Engineers Inc. |
| Seiten | 1-7 |
| Seitenumfang | 7 |
| ISBN (elektronisch) | 9781538627266 |
| ISBN (Print) | 9781538627273 |
| Publikationsstatus | Veröffentlicht - 2017 |
| Veranstaltung | 2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Honolulu, USA / Vereinigte Staaten Dauer: 27 Nov. 2017 → 1 Dez. 2017 |
Abstract
An artificial neural network (ANN) based approach is developed for reconstructing gappy stochastic process records. ANNs can be utilized to capture stochastic patterns in process records in an 'average sense', however individual predictions of missing data can be unreliable. This is both due to the inherent uncertainty introduced by the stochastic process itself, and to uncertainties relating to ANN parameters. Here, instead of relying on single network predictions, an ensemble of networks are trained on the data, the outputs of which are used to estimate statistical distributions for random variables which replace deterministic predictions. In order for this approach to be computationally viable, single-layer feed-forward networks, initialized with random weights are trained via fast, non-iterative least-squares fitting, popularly known as Extreme Learning Machine (ELM) networks. Examples are provided with real data to demonstrate both network prediction effectiveness and uncertainties revealed by varying network parameters.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Artificial intelligence
- Informatik (insg.)
- Angewandte Informatik
- Mathematik (insg.)
- Steuerung und Optimierung
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2017. S. 1-7.
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Revealing prediction uncertainty in artificial neural network based reconstruction of missing data in stochastic process records utilizing extreme learning machines
AU - Comerford, Liam
AU - Beer, Michael
AU - Lu, Naiwei
N1 - Publisher Copyright: © 2017 IEEE.
PY - 2017
Y1 - 2017
N2 - An artificial neural network (ANN) based approach is developed for reconstructing gappy stochastic process records. ANNs can be utilized to capture stochastic patterns in process records in an 'average sense', however individual predictions of missing data can be unreliable. This is both due to the inherent uncertainty introduced by the stochastic process itself, and to uncertainties relating to ANN parameters. Here, instead of relying on single network predictions, an ensemble of networks are trained on the data, the outputs of which are used to estimate statistical distributions for random variables which replace deterministic predictions. In order for this approach to be computationally viable, single-layer feed-forward networks, initialized with random weights are trained via fast, non-iterative least-squares fitting, popularly known as Extreme Learning Machine (ELM) networks. Examples are provided with real data to demonstrate both network prediction effectiveness and uncertainties revealed by varying network parameters.
AB - An artificial neural network (ANN) based approach is developed for reconstructing gappy stochastic process records. ANNs can be utilized to capture stochastic patterns in process records in an 'average sense', however individual predictions of missing data can be unreliable. This is both due to the inherent uncertainty introduced by the stochastic process itself, and to uncertainties relating to ANN parameters. Here, instead of relying on single network predictions, an ensemble of networks are trained on the data, the outputs of which are used to estimate statistical distributions for random variables which replace deterministic predictions. In order for this approach to be computationally viable, single-layer feed-forward networks, initialized with random weights are trained via fast, non-iterative least-squares fitting, popularly known as Extreme Learning Machine (ELM) networks. Examples are provided with real data to demonstrate both network prediction effectiveness and uncertainties revealed by varying network parameters.
KW - artificial neural networks
KW - extreme learning machines
KW - missing data
KW - prediction uncertainty
UR - http://www.scopus.com/inward/record.url?scp=85046089988&partnerID=8YFLogxK
U2 - 10.1109/SSCI.2017.8285295
DO - 10.1109/SSCI.2017.8285295
M3 - Conference contribution
SN - 9781538627273
SP - 1
EP - 7
BT - 2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017
Y2 - 27 November 2017 through 1 December 2017
ER -