It would be interesting to compare our result from the back propagation network with a identical network except to let the all weight changes after one iteration instead of, as today, change them one by one. Some other things to change are the momentum and learning rate, let them be adaptive. And then save all the data and plot it out to some graphs. In this way it would be possible to see how a network is evolving and changes over time. These changes/comparisons would give us a deeper look into the world of neural networks.
One important improvement of our program is the possibility to save the states of each neuron, i.e. all the data of the network. This would make it possible to train the network and later use the trained network in real application. If the program for training and testing the network is very dynamic, maybe because of that not so quick, it gives a good start to use it when developing real applications. To get all the states for each neuron and then easily implement it in a much faster version, for example in assembler or on a silicon chip.
It exists many applications where it would be very interesting to implement a neural network to improve the application. This makes it even possible to make new applications that were not possible before. One example is a system locating human bodies, it could be used in rescue missions, intruder alarms and even in some kind of interactive art systems. Another example is a system for searching and analysing data, for example it could give better search result on the web or finding life in space. Why not, build a system with a web camera and a projector, let the camera survey a room and show the people, just the people, moving around in the room with the projector. This could be an amusing and fascinating contribution to a dreary environment. A further project with a neural network in the central system could be a camera based human body interface to a computer. This would give some very interesting ways of controlling things, for example a robot or computer game. The number and types of applications possible to use an interface like that is enormous.