Limitations on the connection weights due to the introduction of threshold self-adjustment

Application of pseudorehearsal to neural networks with unsupervised learning is an important step in the creating of an artificial memory system with sequential learning. Pseudorehearsal originated from the interaction of two memory subsystems in the human brain, hippocampal and neocortical. The neurons in those two areas differ significantly in their electrical characteristics, but those differences are usually beyond the scope of connectionist modeling. To prove relevance of those parameters to the functionality of the system, the author introduce them in the model. Each of the parameters has its own, non-trivial effect and the best way to study those effects is to do it one by one. The threshold self-adjustment and its relation with the connection weights are studied. The introduction of threshold self-adjustment requires the connection weights to stay within some range to preserve the stability of the net. The suggested modified Hebbian learning rule takes into account those limits.