Long-Term Memory Neural Circuits, Fast and Precise - Brain Theory from a Circuits and Systems Perspective - page 142

Fig. 8.13 Filter example that detects two repetitions for learning

1)] timing filters are needed. A sequence in this
model may go forward or backward. This implies a like number of bus lines for the
filter outputs and a like number of switches S ij (1

In general for K words [K (K

j). Also
there must be K lines for memory word outputs and K lines for memory word
inputs. The interneuron overhead is high, but probably consistent with the com-
plexity of brain circuitry.

i K,1 j K, i 6¼

Savant Learning

Synaptic growth for implicit learning takes excessive time and thus accounts poorly
for photographic memory which can be formed instantly. Special rehearsal filters
may be available to explain rapid learning [ 5 ]. Consider a simplified digital filter
model for rehearsal as in Fig. 8.13 . A toggle pulse occurs when a signal goes from one
particular word to another word, thus passing through the input AND gate U1. Only
one pulse is passed through a weak synapse (triangular symbol) such that T1 toggles.
T2 cannot toggle because by the time T1 toggles, the input pulse is gone. The
AND gate U2 does not permit T1 to toggle until a second occurrence of a signal
from one particular word to another word. After two such signals, the output goes
true (A similar cascade of toggles was suggested in a previous chapter concerning a
recall referee). Thus we have a rehearsal filter example with output after only two
rehearsals, causing an automatic move from one memory word to another.
The output of this rehearsal filter holds via a STM neuron; it holds until long-
term potentiation and a synapse can be developed. “Memory” savants therefore are
proposed to be not memory geniuses, but learning geniuses. They need fewer
rehearsals. This would permit faster learning of sequences, resulting in a long
memorized recitation without conscious effort.
To avoid having a system attempt to learn too much, it may be necessary to
regulate learning using global interneurons; for example, an Enable might be
activated by a savant's motivation to memorize something interesting. Simulated
qubits with rehearsal filters for quick learning in conjunction with quick memori-
zation would explain what is perceived to be phenomenal memory.