Hi Carlos, as PhD student having used ANN, as well as other machine learning approaches, I can tell you that they are becoming, in general, a common solution for a wide range of classification and regression problems. In my case I am using machine and deep learning to assist in the segmentation of medical imaging as well as the prediction of prognosis factors in brain tumors.

In addition, I would like to say that if you are going to start with some topic that you would aim at being pioneering, I would switch to deep learning instead (ANN with more than typical 2 or 3 hidden layers). They offer solutions to the classical problems of shallow architectures.

Cheers

Carlos, 

There are many different models of neural networks. Many neural networks, like MLP and RBF, can be applied in modeling problems, like function approximation, regression and classification (modeling the decision boundary).

SVM, yet another popular NN model, can be applied to solve classification problems. But SVM does not give you the model of the decision boundary.

SOM and LVQ can be applied to solve clustering problems.

Graphical models (e.g. Bayesian networks and statistical learning algorithms) can be applied to solve function approximations, causal relations modeling and classification problems.

Recurrent type neural networks, like RNN, can be applied to model dynamic systems, solve time series prediction, control problems as well.

In the last decade, there are many new problems that have also been solved by neural networks. For example, in biomedical engineering, analyzing the EEG signals is a big problem. The purpose is to discover the regular pattern in the signal when a person is thinking. One application of this results is to let the computer can give electrical signal to control the "robotic legs or arms". As the volume of the signal is huge, many NN researchers have been working on this. Deep learning is one of the model that has been applied to do that job.

Another amazing application of the NN models, graphical model in particular, is mining regular patterns in a social network. People with similar behavior, background, taste, living style etc are found. Recommendations can thus be suggested to the members in the social network. Beyond that, merchants can make use of these finding to do target advertising.

Carlos,

For your information. Some of the current technologies you can see in today are the outcomes of some researches that have been doing for the last two decades. In the 90s, there were hot topics. One is the automatic target tracking problem. One is the auto-parking problem. As you might see, the target tracking technology has been realized in military defense systems. While it is not known what algorithm it is using in the system, the effort the researchers put in the 90s should be credited. About the self-parking technology, it has been implemented in the modern cars. Voice recognition in the cell phone, face recognition in Facebook, are another two amazing features we are using today. They have also been researching for more than two decades.

I still believe that there are a lot of applications out there for neural networks. May be deep learning, may be statistical learning, might be others.

Cheers

Thanks so much for your answers.

Best