We have discrete vector of time series and output that is between [a,b]. Is it probability of occurrence? How we can investigate output? How find what happened there?
A neural network is array of decision making algorithm where combination of neuronal units are used to get a decision out of a series of input. A neuronal unit takes 2 or more input and gives a single output. Combination of units may yield to n number of decisions based on inputs they make.
Neural networks are different from simple programming because, neural network use training sets to train their decision making ability. Unlike normal software algorithms, the decision making pattern changes in neural networks. The decision making precision tends to be more accurate with more and more training datasets used. Once training is finished, the neural networks can be used to predict real data. Sometimes output from real data is used as training datasets after validation.
Example of such kind of neural network is algorithm for Multiple Sequence Alignment.
You should expose the quesiton better. As Gonçalo gutierrez-tobal has well explained, the output of an artificial neural network strictly depends on the natureza of the problem in which you are using it, namely classification and/or regression. Now if you are thinking about the internal mechanism that allows the calculation of the ann output, then you can refer to the saurabh gayali's answer.
It depends on the output function that you specified for the NN and its purpose.
I will assume that you are using it for classification, you have two classes and you trained the net on a data-set with equivalent data vectors per class. You can have a feel for what the output classification is by saying that anything below 0.5 is of class a and anything above 0.5 is of class b then check the accuracy on a testing data-set.
A Neural Network is like a black box that knows how to process inputs to create a useful output. The calculation is quite complex and difficult to understand, yet the results are often useful. These results "outputs" are the appraisal values or updates.
thanks a lot for answer, i have very chaotic and near random data sets and i want NN it to supervised classified it however when i get sign the result after repeated hundred or thousand times is near for example 50.6 correct classified and 49.4 miss classified but if we see output before sign function it is some degree of randomness to and for example if it between [a,b] if divided to any subsets it is probability find each class is near the above answer or sum times the amount of b is related to zero group however it is maximum but it is not related to 1 group how i can find some useful information can i do something else to extract more information? thanks a lot
Your concern/problem was not clear for me. But, if -in general-we are talking about classification problems (or even regression/forecasting) we must remakr that the nature of the model structure is one of the problems we must solve before we get an adequate classificator. Before, we must identify an adequate set of inputs and -in the case of NN-the type of activation functions as well as the interconnections (and initialization, and the in-sample and out-of-sample datasets, the learning algorithm, and so on). So, if the nn you adjusted seems to have a poor performance (almost half/half prob), then you might try to find where the problem is, giving a good look to all the other aspects mentioned.
If it is used for classifying 2 classes, then the output is 1 (for class 1) or 2 (for class 2)
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing