how do i use conditional training in neural network

i am new to matlab and neural networks. i want to train a nn using the condition that are specified. the input is a matrix, and the elements of the output matrix should satisfy specific condition. for eg, the diagonal elements should be zero, the sum of certain elements should be a specified value etc.

also how do i input the matrix. i could only input it as separate rows/columns and not a a whole matrix, whose elements are to be considered together and not individually.

how do i do it using nn toolbox?? should i use the simulink tool boxes??how??

Answer 1

[Greg Heath]

I think you are confused.

The common Feedforward Multilayer Perceptron either implements regression/curvefitting (fitnet) or classification/pattern-recognition (patternnet).

Given an input matrix of N I-dimensional column vectors, it will output a corresponding matrix of N O-dimensional column vectors.

Each output column corresponds to a single inputcolumn. In matrix notation, y = f(x).

However, there are Time-Series nets where the current output depends on

a. current and past values of input (timedelaynet) y(t) = f(x(t-d: t))

b. past values of output (narnet) y(t) = f(y(t-d: t-1))

c. a combination of a and b (narxnet).

Your description does not seem to fit these scenarios.

Hope this helps.

Thank you for formally accepting my answer.

Greg

[remya]

Thank you for answering.

But how do u suggest I approach the problem??

[Greg Heath]

I don't fully understand the general premise. Can you post a small example

1. Explain the desired I/O transformation 2. Post examples of the input matrix and corresponding output target matrix.

Be forewarned that each output column only depends on one input column. If you want an output column to depend on more than one input column, you should stack those columns to form one input column.