| Publikationstyp: | Beitrag in wissenschaftlicher Zeitschrift |
| Art der Begutachtung: | Peer review (Publikation) |
| Titel: | Learning long-term dependencies in segmented-memory recurrent neural networks with backpropagation of error |
| Autor/-in: | Glüge, Stefan Böck, Ronald Palm, Günther Wendemuth, Andreas |
| DOI: | 10.1016/j.neucom.2013.11.043 |
| Erschienen in: | Neurocomputing |
| Band(Heft): | 141 |
| Seite(n): | 54 |
| Seiten bis: | 64 |
| Erscheinungsdatum: | 2014 |
| Verlag / Hrsg. Institution: | Elsevier |
| ISSN: | 0925-2312 1872-8286 |
| Sprache: | Englisch |
| Fachgebiet (DDC): | 006: Spezielle Computerverfahren |
| Zusammenfassung: | In general, recurrent neural networks have difficulties in learning long-term dependencies. The segmented-memory recurrent neural network (SMRNN) architecture together with the extended real-time recurrent learning (eRTRL) algorithm was proposed to circumvent this problem. Due to its computational complexity eRTRL becomes impractical with increasing network size. Therefore, we introduce the less complex extended backpropagation through time (eBPTT) for SMRNN together with a layer-local unsupervised pre-training procedure. A comparison on the information latching problem showed that eRTRL is better able to handle the latching of information over longer periods of time, even though eBPTT guaranteed a better generalisation when training was successful. Further, pre-training significantly improved the ability to learn long-term dependencies with eBPTT. Therefore, the proposed eBPTT algorithm is suited for tasks that require big networks where eRTRL is impractical. The pre-training procedure itself is independent of the supervised learning algorithm and can improve learning in SMRNN in general. |
| URI: | https://digitalcollection.zhaw.ch/handle/11475/15300 |
| Volltext Version: | Publizierte Version |
| Lizenz (gemäss Verlagsvertrag): | Lizenz gemäss Verlagsvertrag |
| Departement: | Life Sciences und Facility Management |
| Organisationseinheit: | Institut für Computational Life Sciences (ICLS) |
| Enthalten in den Sammlungen: | Publikationen Life Sciences und Facility Management |
Dateien zu dieser Ressource:
Es gibt keine Dateien zu dieser Ressource.
Zur Langanzeige
Glüge, S., Böck, R., Palm, G., & Wendemuth, A. (2014). Learning long-term dependencies in segmented-memory recurrent neural networks with backpropagation of error. Neurocomputing, 141, 54–64. https://doi.org/10.1016/j.neucom.2013.11.043
Glüge, S. et al. (2014) ‘Learning long-term dependencies in segmented-memory recurrent neural networks with backpropagation of error’, Neurocomputing, 141, pp. 54–64. Available at: https://doi.org/10.1016/j.neucom.2013.11.043.
S. Glüge, R. Böck, G. Palm, and A. Wendemuth, “Learning long-term dependencies in segmented-memory recurrent neural networks with backpropagation of error,” Neurocomputing, vol. 141, pp. 54–64, 2014, doi: 10.1016/j.neucom.2013.11.043.
GLÜGE, Stefan, Ronald BÖCK, Günther PALM und Andreas WENDEMUTH, 2014. Learning long-term dependencies in segmented-memory recurrent neural networks with backpropagation of error. Neurocomputing. 2014. Bd. 141, S. 54–64. DOI 10.1016/j.neucom.2013.11.043
Glüge, Stefan, Ronald Böck, Günther Palm, and Andreas Wendemuth. 2014. “Learning Long-Term Dependencies in Segmented-Memory Recurrent Neural Networks with Backpropagation of Error.” Neurocomputing 141: 54–64. https://doi.org/10.1016/j.neucom.2013.11.043.
Glüge, Stefan, et al. “Learning Long-Term Dependencies in Segmented-Memory Recurrent Neural Networks with Backpropagation of Error.” Neurocomputing, vol. 141, 2014, pp. 54–64, https://doi.org/10.1016/j.neucom.2013.11.043.
Alle Ressourcen in diesem Repository sind urheberrechtlich geschützt, soweit nicht anderweitig angezeigt.