You can start with an easy but VERY powerful version of NN: ELM. Here is my paper about it, you can find all the necessary references in it: http://dx.doi.org/10.1109/TNN.2009.2036259
Stick to the classics: the Multilayer Perceptron with Backpropagation Here's a lecture on it, but there are probably thousands of Youtube and presentations http://www.cs.bham.ac.uk/~jxb/NN/l7.pdf
i want to apply multilayer perceptron in structural engineering, In tall building design process. I study some books and paper related ANN but i am still confuse can MLP give accurate result if data are very small. about 50 data of input variable 20 and output variable 4 are available with me, please suggest me .. i study about K -fold cross validation it can solve my problem.
actually i am student of structural engineering field i am poor in matlab suggest me can i get any sample M file for multilayer perceptron with k fold cross validation please suggest me it help me to understand,([email protected])
Hi Lila. A neural network will be suitable if your problem is a nonlinear one, i.e. you need to approximate a complex nonlinear relationship. If you have only 50 observations of 20 variables, but they have little noise and adequaely capture the relaionship, this might easily be enough data. If you have more noise, or the data is concentrated on some (non-representative) regions of feature space or dispersed all over it then it might not be as simple ... so explore the data (plot it) and check fro nonlinearities and interactions, and then try a few neural net approaches.
If you like video tutorials, i recommend to you watch the Andrew Ng's Machine Learning course on coursera.org specially week 4 and week 5 that dedicated to neural networks.
You can found it:
https://class.coursera.org/ml-2012-002
If you wanna have a deep understanding and need advanced topics, i recommend to you that watch Hinton's Neural Networks for Machine Learning in coursera.org.