If I have a set of input data and a set of output data as shown in attached file,
Can a Neural Network predict an output if the input data out of the given input data?
Example: if the new input is (28,42) ====> what is the output?
If yes, What if the given data sets for input and output are random? Not regular data? Can NN predict the output?
Hi,
you are fortunate today :) because i have a similar work in kaggle that i can share with you which will clarify your confusion with ease,here is my kaggle kernel : https://www.kaggle.com/mobassir/hello-world-of-neural-network
you need to understand that clearly,the code i have written there uses python,i hope you understand python,if not then it doesn't matter much,let me explain how it works :
by analyzing provided data a neural network learns the mathematical definition which describes the relation between input to output mapping of your data,for example the data you've provided here has a mathematical definition which is : y = x2-x1
but if the given datasets for input and output are random or Not regular data then NN's will make mistakes just like we human do for example i teach you that 1+1 = 5,2+2 = 3.5,100+45 = -108.6
then what will you answer me if i ask you that : 6+33= ?
you'll make mistake because there is no good pattern in the input set that you learnt well to predict the outcome of 6+33.all you did is memorized the samples i have provided because you didn't understand that if 1+1 = 5 then why 2+2 = 3.5 and so on,so what you did is just memorized the samples and when i ask you new question, you fail to answer me.
similarly, your neural network tries hard to find pattern from your dataset through changing weights with forward and backward propagation so if the dataset is random which doesn't have good patterns to describe the X to y mapping then definitely your neural network will eat dirty and we machine learning enthusiasts call it overfitting
Of course a neural network can be constructed to do so; but it's necessary to understand what this means. It means that, based on the available data, the neural network can be found that will try to represent the rule they express and produce an output that's consistent with that rule.
One way such a network can be found is the tiling algorithm: http://www.lptms.u-psud.fr/membres/mezard/Pdf/89_MN_JPA.pdf
Once the architecture has been found, that can represent the rule at all, there are many ways to explore how well the new information is consistent with the rule. One way to do this is presented here: Article Stability-capacity diagram of a neural network with Ising bonds
Hi,
you are fortunate today :) because i have a similar work in kaggle that i can share with you which will clarify your confusion with ease,here is my kaggle kernel : https://www.kaggle.com/mobassir/hello-world-of-neural-network
you need to understand that clearly,the code i have written there uses python,i hope you understand python,if not then it doesn't matter much,let me explain how it works :
by analyzing provided data a neural network learns the mathematical definition which describes the relation between input to output mapping of your data,for example the data you've provided here has a mathematical definition which is : y = x2-x1
but if the given datasets for input and output are random or Not regular data then NN's will make mistakes just like we human do for example i teach you that 1+1 = 5,2+2 = 3.5,100+45 = -108.6
then what will you answer me if i ask you that : 6+33= ?
you'll make mistake because there is no good pattern in the input set that you learnt well to predict the outcome of 6+33.all you did is memorized the samples i have provided because you didn't understand that if 1+1 = 5 then why 2+2 = 3.5 and so on,so what you did is just memorized the samples and when i ask you new question, you fail to answer me.
similarly, your neural network tries hard to find pattern from your dataset through changing weights with forward and backward propagation so if the dataset is random which doesn't have good patterns to describe the X to y mapping then definitely your neural network will eat dirty and we machine learning enthusiasts call it overfitting
Definitely yes. It is the main profit of an ANN.
About the second question, you need to know that ANN creates some learning rules for itself and applies them to the future data (unseen input values). So, if your new data are completely stranger with those have been used for training the network, it can not be expected to perform excellently. Please notice that I'm talking about the range of the data not their sequence.
Hi Amr
Based on your input and output. For question 1, if you input [2 ,3] you will get [1] as output. If you input [28, 42], you will get [14] as answer.
For question 2, neural network will still provide a answer even if you use a random input which does not have the similar input and output relationship as the dataset. However the result won't be as accurate.
How can predicted output value be determined using MATLAB Neural network toolbox?
Hi Umair Shahzad
You need to train the ANN first, after completing all the training process you can get a MATLAB function which contains the trained ANN. Enter the samples (same structure as your input) to the MATLAB function, the output of the function is the predicted value.
- Badges
- Science topic
- Similar topics
- Filing