When we want to train neural network, we found at least three parameters like input, target and output. I do not know what the target is and how it could be selected.
Target, T, is the desired output for the given input, X.
Train the network with known input (X) and target (T).
The output of the resulting design, given the input, is output , Y.
The error is e = T-Y. Of course the most common ultimate goal of training is to minimize the mean-squared-error.
If you are lucky enough to have quantifiable Target in the first place :) and to have this quantified target adequately represented among the event attributes, then yes - the target is a part of the output.
The difference between input and output is not categorical. It exists just in Time separating attributes that are known in the decision point from other attributes (including the target) which belong to the prediction.
Target is the "correct" or desidered value for the respose associate to one input. Usually, this value will be compared with the output (the response of the neural network) to guide the learning process involving the weight changes.
For a particular set of input ('s) there will be some actual output response (like output of experiment for some set of input parameters.). While creating neural network using nntool in Matalb, it is required to provide input data as well as output data in form of vector ('s). Then one can create required ANN structure by choosing proper algirithm, number of input and outout layers and nuerons, and by selecting transfer functions.
Target, T, is the desired output for the given input, X.
Train the network with known input (X) and target (T).
The output of the resulting design, given the input, is output , Y.
The error is e = T-Y. Of course the most common ultimate goal of training is to minimize the mean-squared-error.
Simply put, if you are given a data-set, target is the ground-truth.
target means your Ideal Output
But generally we not find ideal output.
and difference of ideal and real output is error
and this is your work is by using any technique reduce this error
Target is to reach the weights(between neural layers) by which the ideal and desired output is produced.
The target is to find the weights that perform the black magic of classifying new data using information of an almost 'perfect' classification done during training (by minimising an error function). It does its job, but essentially it is a black box, hence I don't like it.
From Abdul's apt diagram, the target is simply the result/answer/value you are trying to match your output with which then informs/guides your weighting process. The target provides a guide as to how your weights should be adjusted to have the least difference possible between your target and output.
What is the beneficial of having a neural network in a controller, when we are training the network to follow the controller target?
It is your actual output data. Neural network tries to match its predicted data with this target and then you try to minimize the difference between the two.
- Badges
- Science topic