Researches of the mechanisms of sequence acquisition by hand movement allow us to propose the hypothesis about hemispheric-specific coding mechanisms in human memory – positional coding (hand’s working points) and vector coding (hand’s movements) (Lyakhovetskii, Bobrova, 2009; Bobrova et al., 2011, 2012). This hypothesis is reinforced by both neurophysiological (Grafton et al., 1992) and clinical (Harrington, Haaland, 1991) data, and by neural network model qualitatively reproducing different characteristics of memorization (primacy-effect, errors distributions, exchange errors) (Lyakhovetskii, Bobrova, 2009; Lyakhovetskii, Potapov, 2012; Lyakhovetskii et al., 2012). Is it possible, using the same model, to simulate processes of learning at repeatable reproduction of the same sequence?
The weighting matrix of heteroassociative network, storing the pairs of sequence’s elements, was changed during learning with the help of QLBAM algorithm. The learning was performed without feedback during psychophysiological experiments – the cycles of reproduction of memorized movements of volunteer’s right or left hand by 6 positions at A4 sheet were repeated. As simulation's criterion of learning the number of iterations for convergence to steady state was chosen. This information is available for model during unsupervised learning. The number of model’s learning iterations coincides with the number of volunteers’ learning iterations.
The 10000 model experiments were performed with the same stimuli, which were used in psychophysical experiments. It is shown that the model qualitatively reproduced the form of psychophysical curve using positional coding – gradual increase of performance up to threshold during learning. Despite of absence of a priori information about different errors types, the model is able to reproduce the ratio of durations of correct/wrong responses: the duration of correct responses is lower than the duration of responses with repeatable errors that in turn is lower than the duration of responses with other errors. The increase of performance both for model and for volunteers occurs first of all owing to decrease of the other errors and only after that owing to decrease of the repeatable errors. Using of vector coding during the model's learning with the help of QLBAM algorithm doesn’t lead to increase of performance that contradicts psychophysical data. Thus, the results allow to assume, that right and left hemisphere of human brain uses different algorithms during learning of movements’ sequences.