Hi Ehsan,

The performance of a particular ANN directory depends on the problem. There is a direct correlation between the performance of your ANN and the feature extraction (how well you have extracted the features, so that the ANN can significantly differentiate them during the training process). Also, you have to choose a suitable ANN structure which suits your problem. 

Though there are many new and old ANN structures, you would have to choose among them the best which suits your problem.

Regards 

I recommend the Bayesian regularization back propagation neural network

Just some resources for deep learning if you are interested.

http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial

Courtesy to Dr. Andrew Ng.

Hi Ehsan,

The performance of a particular ANN directory depends on the problem. There is a direct correlation between the performance of your ANN and the feature extraction (how well you have extracted the features, so that the ANN can significantly differentiate them during the training process). Also, you have to choose a suitable ANN structure which suits your problem. 

Though there are many new and old ANN structures, you would have to choose among them the best which suits your problem.

Regards 

It is acutally based on your problem and the used feature extractor. Still, I agree with Asta that Deep Learning is a powerful learning algorithm

I agree with "based on the problem" to some extent. Actually, some problems need much complex, such as recurrent NN, while others needs just some kind of feedforward. However, two main state-of-the-art are Spiking NN and Deep NN.