Computational memory recall. The resultant 350
units network was further evaluated by using a computational
recall procedure. In the recall procedure, a scene unit with a
receptive field covering all grids (with unit number 45) was
activated initially. Then other CA3 units were automatically
activated according to the stored recurrent connections. The
global inhibition of the CA3 activation further supports the
sequential activation of the CA3 units. The recalled sequence was
decoded by using input visual features in terms of temporal coding.
The number of successfully recalled object-place associations, R,
was used as a predictor of subject’s subsequent recall. Equations
and parameters of the recall procedure are described in Text S2.