Training feedforward networks with the Marquardt algorithm
Обучение прямых (feedforward) нейронных сетей с использованием алгоритма Марквардта
1994-01-01
SCID: 54.1/3exrf8w9
Discuss with AI
Marquardt algorithmbackpropagationconjugate gradientfeedforward neural networksfunction approximationnonlinear least squarestraining efficiencyvariable learning rate
Figures from the paper
Abstract (AI)
The Marquardt algorithm for nonlinear least squares is presented and is incorporated into the backpropagation algorithm for training feedforward neural networks. The algorithm is tested on several function approximation problems, and is compared with a conjugate gradient algorithm and a variable learning rate algorithm. It is found that the Marquardt algorithm is much more efficient than either of the other techniques when the network contains no more than a few hundred weights.
Key Findings
1
The Marquardt algorithm (for nonlinear least squares) can be incorporated into backpropagation to train feedforward neural networks.
2
The Marquardt algorithm is much more efficient than conjugate gradient and variable learning rate methods for networks with up to a few hundred weights.
3
The efficiency advantage is reported specifically for networks containing no more than a few hundred weights.
4
When tested on several function approximation problems, the Marquardt-trained networks were compared to conjugate gradient and variable learning rate algorithms.
Research Object
Feedforward neural networks trained using backpropagation
Research Subject
Effectiveness and efficiency of the Marquardt algorithm (nonlinear least squares) for training feedforward networks, compared to conjugate gradient and variable learning rate methods, especially for networks with up to a few hundred weights
Publication Details
Publication Date
1994-01-01
Journal
Publisher
ISSN
Access Type
Author Information
Download PDF
Subscribe to digest