A fast learning algorithm of neural network for the training and recognition of the phonemes

In order to improve the training speed of multilayer feedforward neural networks, we proposed and explored fast backpropagation (BP) algorithms by introducing the hybrid global optimization conjugate gradient algorithm for the dynamic learning rate. This was to overcome the BP learning problem which...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:ISIMP 2004 : proceedings of 2004 International Symposium on Intelligent Multimedia, Video, and Speech Processing : October 20-22, 2004, Hong Kong s. 318 - 321
Hlavní autoři: Minghu Jiang, Hongmei Pang, Beixing Deng, Chengqing Zong
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 2004
Témata:
Automation
Computational linguistics
Content addressable storage
Data engineering
Gradient methods
Iterative algorithms
Multi-layer neural network
Natural languages
Neural networks
Pattern recognition
ISBN:9780780386877, 0780386876
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:In order to improve the training speed of multilayer feedforward neural networks, we proposed and explored fast backpropagation (BP) algorithms by introducing the hybrid global optimization conjugate gradient algorithm for the dynamic learning rate. This was to overcome the BP learning problem which caused plunging into local minima or slow convergence. Our algorithm is of a higher recognition rate than that of the Polak-Ribieve conjugate gradient and conventional BP algorithms. It showed less training time, less complication and stronger robustness than the Fletcher-Reeves conjugate gradient and conventional BP algorithms for real speech data.
ISBN:9780780386877
0780386876
DOI:10.1109/ISIMP.2004.1434064