Based on the 2005 textbook Introduction to Neural Networks Using MATLAB 6.0
| Old MATLAB 6.0 (PDF) | Modern MATLAB (2024) | Explanation | | :--- | :--- | :--- | | newff(minmax(P), [5 1], 'tansig' 'purelin', 'trainlm') | feedforwardnet([5 1]) | The architecture is now encapsulated in feedforwardnet . | | train(net, P, T) | net = train(net, P, T) | You must assign the output back to the network. | | sim(net, P_test) | net(P_test) | You can now call the network as a function directly. | | init(net) | net = init(net) | Similar assignment requirement. | | learnbp (manual backprop) | Obsolete; use train with 'traingd' | The toolbox has automated this. | introduction to neural networks using matlab 6.0 .pdf
Even in 2000, the concepts of overfitting and generalization were critical. The PDF will explain how MATLAB 6.0 split data, how to use train to iterate through epochs, and how to plot the mean squared error (MSE) using plotperf . Based on the 2005 textbook Introduction to Neural
The book does a fantastic job explaining why RBFs are faster than backprop for function approximation. | | train(net, P, T) | net =