GA is an optimization method and ANN is a training method. One of the use of GA is to optimize the weights of ANN. There are some tools available where we train the ANN from given data and GA is used as an optimization algorithm. One of the tool I have used is 4CastXL. This tools uses three optimization methods: (i) Backpropagation (ii) Genetic Algorithm and (iii) Conjugate Gradient.
Other use of GA in ANN is to optimize number of hidden nodes. As we know that number of hidden layer and hidden nodes play an important role in training time as well as Mean Square Error (MSE) so it becomes important to decide number of hidden nodes using optimization algorithms like GA.
GA can be used to identify the optimal set of connection weights in the network.
I'm not too familiar with the literature on neural networks, but it wouldn't surprise me if someone had used GA to actually *construct* the neural network, and then train it afterwards (optimal topology + optimal weights).
GA is an optimization method and ANN is a training method. One of the use of GA is to optimize the weights of ANN. There are some tools available where we train the ANN from given data and GA is used as an optimization algorithm. One of the tool I have used is 4CastXL. This tools uses three optimization methods: (i) Backpropagation (ii) Genetic Algorithm and (iii) Conjugate Gradient.
Other use of GA in ANN is to optimize number of hidden nodes. As we know that number of hidden layer and hidden nodes play an important role in training time as well as Mean Square Error (MSE) so it becomes important to decide number of hidden nodes using optimization algorithms like GA.
Mahesh, I am curious as to what objective function you use to establish an optimal topology for the artificial neural network. Is it still the mean square error of the prediction set (I forget the correct term for it; the data set you use to verify that the ANN is working correctly post-training)? Or perhaps there is some statistical hypothesis one can use to establish that an optimal topology has been found?
For optimizing number of hidden nodes, search space is less as compared to number of weights but still GA can be used effectively for optimizing number of hidden nodes.
MSE can be used as fitness function for GA to optimize number of hidden nodes.
Neural network can be used as an objective function of GA. In practical problems where we are unaware of the actual relation between the variable and the output (mathematically), we can generate an ANN (Artificial Neural Network). Which then drives the genetic algorithm.
Thus one can use ANN and GA in coupled mode for an optimization problem.
Genetic Algorithm is an optimization technique, and can be used for any type of optimization problem such as single objective or multi-objective. As far as application of genetic algorithm for neural network is concerned, I will recommend a paper title given below:
"A multiobjective genetic algorithm for obtaining the optimal size of a recurrent neural network for grammatical inference"
Although this paper is given for grammatical inference problem, but you can apply GA depending on the problem you are dealing with for neural network. The paper will guide you.
I think GA used for training the NN ,in other word it try to optimize the connection weights to reduce the time ,but we need how to build and coding the problem.