Figure 7.
Architecture of the hippocampal module after the exploration phase in the example of episodic memory retrieval. These numerical results of learning in the hippocampal module were achieved in a typical memory-exploring sequence of 10 moves (Supplemental online material at http://www.krasnow.gmu.edu/ascoli/sa_lm05_som/: the movie “exploration”). Labels A–E within the circles correspond to the hippocampal pointers in Figure 5 and the related stable neocortical patterns in Figure 6. Solid CA3-to-CA3 arrows (bottom) show explored (and stored in the hippocampal module) transitions among memories in the neocortical component. Dashed CA3-to-CA3 arrows show available, yet unexplored transitions. Despite the unexplored part of connections, all-to-all off-diagonal synaptic weights W (vertical oblique arrows of variable thickness) were modified by learning according to equation 1 in this simulation. The thickness of each arrow is proportional to the corresponding computed synaptic weight W, which varies from 0.2 to 1.0. Straight vertical CA3-to-CA1 dotted lines that represent diagonal matrix elements had the maximal weight Wii = 2.