IEICE TRANSACTIONS on transactions
Learning Temporal waveforms in Neural Networks
Summary :
An approach is described to synthesis and recognition of temporal patterns by using neural networks. A neural network is trained to produce prescribed waveforms with the steepest descent method which optimizes analog dynamics of neural networks described by differential equations. First a technique is developed for calculating error sensitivities with respect to network parameters by the adjoint network approach. Next an upper bound on timesteps is established to ensure the stability of the numerical solutions of the differential equations of networks. The effectiveness of these techniques are verified by several examples of learning of transient or oscillating waveforms with simple networks. In addition the complexity of the waveform is discussed which can be synthesized by a simple class of neural networks.
- Publication
- IEICE TRANSACTIONS on transactions Vol.E73-E No.12 pp.1925-1931
- Publication Date
- 1990/12/25
- Publicized
- Online ISSN
- DOI
- Type of Manuscript
- Special Section PAPER (Special Issue on the 3rd Karuizawa Workshop on Circuits and Systems)
- Category
- Neural Networks
Authors
Keyword
Latest Issue
Copyrights notice of machine-translated contents
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Cite this
Copy
Kiichi URAHAMA, "Learning Temporal waveforms in Neural Networks" in IEICE TRANSACTIONS on transactions,
vol. E73-E, no. 12, pp. 1925-1931, December 1990, doi: .
Abstract: An approach is described to synthesis and recognition of temporal patterns by using neural networks. A neural network is trained to produce prescribed waveforms with the steepest descent method which optimizes analog dynamics of neural networks described by differential equations. First a technique is developed for calculating error sensitivities with respect to network parameters by the adjoint network approach. Next an upper bound on timesteps is established to ensure the stability of the numerical solutions of the differential equations of networks. The effectiveness of these techniques are verified by several examples of learning of transient or oscillating waveforms with simple networks. In addition the complexity of the waveform is discussed which can be synthesized by a simple class of neural networks.
URL: https://global.ieice.org/en_transactions/transactions/10.1587/e73-e_12_1925/_p
Copy
@ARTICLE{e73-e_12_1925,
author={Kiichi URAHAMA, },
journal={IEICE TRANSACTIONS on transactions},
title={Learning Temporal waveforms in Neural Networks},
year={1990},
volume={E73-E},
number={12},
pages={1925-1931},
abstract={An approach is described to synthesis and recognition of temporal patterns by using neural networks. A neural network is trained to produce prescribed waveforms with the steepest descent method which optimizes analog dynamics of neural networks described by differential equations. First a technique is developed for calculating error sensitivities with respect to network parameters by the adjoint network approach. Next an upper bound on timesteps is established to ensure the stability of the numerical solutions of the differential equations of networks. The effectiveness of these techniques are verified by several examples of learning of transient or oscillating waveforms with simple networks. In addition the complexity of the waveform is discussed which can be synthesized by a simple class of neural networks.},
keywords={},
doi={},
ISSN={},
month={December},}
Copy
TY - JOUR
TI - Learning Temporal waveforms in Neural Networks
T2 - IEICE TRANSACTIONS on transactions
SP - 1925
EP - 1931
AU - Kiichi URAHAMA
PY - 1990
DO -
JO - IEICE TRANSACTIONS on transactions
SN -
VL - E73-E
IS - 12
JA - IEICE TRANSACTIONS on transactions
Y1 - December 1990
AB - An approach is described to synthesis and recognition of temporal patterns by using neural networks. A neural network is trained to produce prescribed waveforms with the steepest descent method which optimizes analog dynamics of neural networks described by differential equations. First a technique is developed for calculating error sensitivities with respect to network parameters by the adjoint network approach. Next an upper bound on timesteps is established to ensure the stability of the numerical solutions of the differential equations of networks. The effectiveness of these techniques are verified by several examples of learning of transient or oscillating waveforms with simple networks. In addition the complexity of the waveform is discussed which can be synthesized by a simple class of neural networks.
ER -
English
