The Trainable Model of Memorization of Movement Sequence Based on Heteroassociative Neural Network

The bidirectional heteroassociative neural network model is developed on the basis of psychophysical experiments studying processes of memorization of the hand movements’ sequence. The model qualitatively reproduces the different characteristics of human errors. The learning processes are simulated with the help of QLBAM algorithm. The revealed different classes of human errors — repetitive and other errors have different timing characteristics. They are associated apparently with different stages of the process of sequence learning. The changes of different types of model errors during training and the number of iterations required to the model for transition to a stable state are similar to the such changes in the quantities of psychophysical experiments. It is assumed that the effect of keeping the repeated errors in psychophysical experiments can be interpreted in the model as the preservation of network false attractor in memorizing. Simulating of learning at memorization using QLBAM algorithm allows to reproduce the reduction of non-repeated errors that was demonstrated in psychophysical experiments. Thus, at the first stage the linear algorithm, forming a symmetrical matrix having a lower capacity, is used. At the second stage the QLBAM algorithm iteratively solving nonlinear optimization problem, producing an asymmetric matrix of greater capacity, is used. It can be assumed that these stages reflect the processes of human memorization of movements sequences in working and long-term memory.