Neural Networks (NN), Artificial Intelligence (AI) and other such things could be our future and this is
opinion not only of some of SF authors. This do not have to be true of course. Anyway, we should be prepared...
This experiment is a part of bigger project realised in the Chair of Modelling the Teaching Processes at the Faculty of Physics and Applied Informatics, University of Łódź, Poland. The leader of the project is Tadeusz Wibig (e-mail: firstname.lastname@example.org, www: http://zpk.u.lodz.pl/~wibig).
In this e-experiment We wish to test the architecture of artificial neural network (ANN)
which is able to solve the complex, noisy problems.
We would like to compare the ANN performance with the human (Natural) Neural Network (NNN).
To make the competition trustful we should use some abstract pattern input and add a random
noise to make the classification not obvious.
We used (10 x 10) pixel matrix with the pattern of horizontal, vertical, and two inclined (parallel to both diagonals) thick lines. The empty pattern are also present (with no thick line) to complicate the problem. We wish the network to classify every given pictures to one of five classes. If the noise is relatively small the problem is easy (for ANN as well as for NNN). On the pictures to the left the relatively clear examples are given.
When noise level increases the problems arise in some cases, as it is shown below:
When the noise level gets higher, classification gets harder.
Closer inspection of the figures shows that at the right lower corner there are four small squares with the one sometimes filled black. This is the
The set of pictures above is similar to the one which will be a subject of out test. Please, look at them carefully, compare your guess with the expected, correct answer, before you go to the test on the next page.