Neural Network Backpropagation Optimization

27 次查看（过去 30 天）
Lask
Lask 2019-5-13
编辑： Lask 2019-5-14
Hi all,
I am training a shallow neural network with the following layers using the Neural Network Toolbox.
input layer - 100 (features)
hidden layer - 10 (neurons)
output layer - 100 (labels)
I used the following code to format the training set.
[inputs,~] = tonndata(features,1,0);
[labels,~] = tonndata(targets,1,0);
so, after these two lines, inputs and labels are two 1x5000 cell arrays, being each element in the arrays a 1x100 signal. Then I configure, initialize and train the network using Levenberg-Marquardt backpropagation like this:
net = configure(net,inputs,labels);
net = init(net);
[trained_net,perf] = train(net,inputs,labels);
the network view is as follows:
the issue is, during training, each epoch takes around 2 minutes to train, which seems to much. I know Levenberg-Marquardt algorithm involves the calculation of the inverse of the Jacobian, which takes time, but MATLAB's documentation literally says:
"trainlm is often the fastest backpropagation algorithm in the toolbox, and is highly recommended as a first-choice supervised algorithm, although it does require more memory than other algorithms."
I would like to know if anyone has any suggestion to reduce the training time or if I am missing something in the model setup.
Thanks in advance.

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by