||Licensed User Center|
NeuroSolutions for MATLAB toolbox vs MATLAB's Neural Network ToolboxThere are many differentiating factor between MATLABís own neural network toolbox and NeuroSolutions for MATLAB toolbox. One of them is the ease-of-use built into the commands for using the functionalities available in the toolbox.
To illustrate, let us see how to create a new neural network, say a simple feed-forward network, using the two toolboxes. The commands shown below are written in their simplest form for better understanding.
MATLABís neural network toolbox:
p = minmax(x); % x is the input data
NeuroSolutions for MATLAB neural network toolbox:
mynet = nsGFF;
Firstly, it can be seen that for initializing a neural network in MATLABís neural network toolbox needs to know the min and the max of the input data x and also the number of neurons or processing elements in the different layers of the neural network. Whereas the NeuroSolutions for MATLAB command does not require any information about the input data or any other information about the topology of the neural network.
There would also be some question in a beginnerís mind as to how many layers must there be in a neural network and would the input and the output be considered as layers in the architecture. In MATLABís neural network toolbox the inputs are not considered as a layer and the outputs are considered as a layer. There would also be some doubt as to what the right setting would be for each layer. Should it be 30 neurons or 3 in a particular layer? How will it impact your results?
It will take quite some time for a beginner to understand the intricacies involved in building a neural network. The NeuroSolutions for MATLAB toolbox in turn offers a solution whereby you would not have to spend many taxing hours on learning the product, but would rather spend the time on solving your problem on hand.
There are smart modules built into the interface, which compute all the necessary parameter values on its own. When the parameter values are dependent on the data they are computed whenever the data is first encountered. Hence the number or neurons/PEs* are computed automatically and so are the normalization coefficients, step size etc., In cases where this cannot be done NeuroSolutions for MATLAB sets those parameters to well-researched values. Therefore the user does not have to know a lot about neural networks or about the product to begin using it.
What about Advanced Users?
NeuroSolutions for MATLAB offers a lot of flexibility for advanced users who want to get the most out of their neural networks models. The NeuroSolutions for MATLAB toolbox is integrated with NeuroSolutions, which is the industry standard solution for neural network software. NeuroSolutions offers the users virtually unlimited freedom to build and modify neural models. Itís easy-to-use illustrative graphical user interface has captured the minds of thousands of users worldwide for many years now. Neural models can be built in NeuroSolutions and used in MATLAB through the NeuroSolutions for MATLAB interface. This NeuroSolutions for MATLAB toolbox offers the power of NeuroSolutions neural networks inside MATLABís computing environment thereby reducing your neural network development time by half.
Some other useful functionalities not available in MATLAB's neural network toolbox
To find help for a particular parameter in the neural network object, say the 'numInputs', in the MATLABís neural network toolbox. There is no simple way to understand what 'numInputs' correspond to and how to set them appropriately. Whereas you can get help for any parameter in the NeuroSolutions for MATLABís neural network object by typing .help after the parameter name.
The output of the neural network and the desired data can be compared visually by just setting a parameter to true. The plot will be plotted automatically after training.
mynet.OutputAndDesired = true;
Generate Code using Demo
If you have data loaded into the MATLAB workspace, then it can be used inside Demo 4 ("Demo 4: User your own data"), by typing 'nsDemos', to generate code that will create the neural model, modify appropriately and train the model on your data. The demo will also show how well the neural network was trained along with the output and desired plot. It is a great way to get started on your problem.