Improvement of associative memory by means of inductive learning

Artificial neural networks acquire their average knowledge by learning a huge number of instances. However, in the real world, there are many instances which can not be generalized by such a learning method. The neural network constructed by the conventional learning method is not able to recognize a new instance, which is against the average knowledge. On the other hand, inductive learning in an artificial intelligence constructs knowledge representing the new instance as an exceptional knowledge and can recognize it well. In this paper, we first show that the correct recognition ratio increases as the number of training data increases. Next, we attempt to improve the associative memory by constructing an exceptional instance knowledge when an unknown instance is given. We apply the proposed method to facial expressions recognition in order to confirm the advantage of it.

[1]  T. Takagi,et al.  Recognition of facial expressions using conceptual fuzzy sets , 1993, [Proceedings 1993] Second IEEE International Conference on Fuzzy Systems.

[2]  Patrick Henry Winston,et al.  Learning structural descriptions from examples , 1970 .

[3]  P. Ekman Unmasking The Face , 1975 .

[4]  B Kosko,et al.  Adaptive bidirectional associative memories. , 1987, Applied optics.