I am not able to understand the question. How can we train Neural Network using SVM?
Per my understanding for training Neural Network, we use gradient optimizers to optimize the weights. Are you thinking of using SVM to find optimized weights of Neural Network? Please help me understand the requirement
Your question is not quite clear, so I'll suggest two scenarios:
1- You can take the training data and derive different training sets from it. Then, build a prediction model from each one using the desired methods (i.e., Neural Network and SVM). Finally, combine them to produce an ensemble of learned models.
2- Use SVM as an evaluator to the selected features (i.e., assuming that you need to perform attribute selection). After that, Neural Network constructs the prediction model from those features.
SVM and NN are both supervised learning methods, but they work a bit differently.
Support Vector Machine: SVM fits a hyper plane/function between 2 different classes given a maximum margin parameter.
Neural Network: A neural network has several input, hidden, and output nodes. Each node applies a function some data (could be soft-max, linear, logistic), and returns an output.
The big difference between the two methods is that stochastic gradient descent isn't guaranteed to find the optimal set of parameters when used the way NN implementations employ it. However, any decent SVM implementation is going to find the optimal set of parameters. People like to say that neural networks get stuck in a local minima while SVMs don't.
It might utilize from many layer NN and have the final classification via SVM at the output layer. It is likely to have better classification results compared to normal NN.
You can use Matlab Classification Tool as well as easiest way is to use weka which is just drag and drop. You do not need to play with any codes in weka.