Information Technology Reference
In-Depth Information
Fig. 8.12 Model for learning; three word state machine within subconscious long-term memory
Figure 8.12 attempts to depicts a small-scale subcircuit that is capable of fast
state machine learning, stepping from any given memory word to any other
memory word within a given region of subconscious long-term memory. The
Filters (Flts) in the figure detect repetition and subsequently hold a signal that
enables synaptic growth. Synapses (S ij ) are modeled as field effect transistors
serving as switches in the figure. Interneurons underlie the biological equivalent
of this circuit. Interneurons may hold learning connections and release
neurotransmitters to foster synaptic growth, thus supporting the S ij in this circuit
model for sequential learning.
The design in the figure implies that a system may learn to perform state
transitions from any given word to any other given word in a finite block of K
words within long-term memory. Figure 8.12 shows three words, although more are
possible. The theoretical number of possible sequences is K!, which is rather large,
but by posing simple restrictions, such as having only ascending sequences, this
number is greatly reduced.
For example, consider the sequence 1-2-3 that has been used, say, ten times so
that it is well learned. Then, whenever Word 1 is addressed (associatively) by STM,
after a short delay, Word 2 is addressed directly by Word 1. The path of the signal is
indicated by the dotted line in the figure. Then after another short delay, Word 3 is
addressed directly by Word 2. This will accomplish a three-step procedure without
conscious involvement involving cue editing and recall refereeing.
Note that switches S ij (1
j) enable the state machine.
For example, switch S 12 applies a done signal from Word 1 to the delayed enable
of Word 2. Switches are implemented biologically with synaptic connections
promoted by interneurons carrying serotonin and possibly other synapse-promoting
neurotransmitters.
i
K,1
j
K, i
6ΒΌ
 
Search WWH ::




Custom Search