Information Technology Reference
In-Depth Information
simulation was designed to compare the mutual association abilities between conven-
tional MCNN and Amygdala - Hippocampus model. This simulation was also de-
scribed in [10], however, new statistical results would be reported more in this paper.
The second simulation was designed to compare one-to-many association abilities,
and long-term memory formation abilities between Amydala-Hippocampus model
and the model of limbic system proposed in this paper.
3.1 Simulation of Mutual Association
Two time series patterns used in mutual association simulation of MCNN and Amyg-
dala-Hippocampus model are as same as those in [7] and [10]. All parameters value or
their initial value used in these simulations were decided by empirical knowledge or
according to the previous works and they are shown in Table 1. The comparison of
storage time and recollection time of different models is shown by Table 2.
Amygdala-Hippocampus model had a faster storage than MCNN however a slower
recollection. The results can be explained as that emotional model enhanced the stor-
age processing in the meaning of efficient and enhanced the recollection more “care-
fully” in the meaning of precision. In fact we confirmed that retrieved patterns using
emotional control method showed their completeness meanwhile MCNN failed some-
time. Illustrations of the results by MCNN are omitted here for want of space, and
results by Amygadala-Hippocampus model are shown in Fig. 2.
3.2 Simulation of Long-Term Memory
Two time series patterns shown in Input layer of Fig. 3 similar to the simulations of Ito
et
al
's [5] were used to investigate the association and long-term memory formation abili-
ties of proposed model of limbic system. Binary patterns in each time series were or-
thogonal, and a 4-step interval between the two time series was set to distinguish them.
The procedure of simulation is described according to time sequence as following:
A. Input the two time series patterns. The first pattern of each time series was
same to serve as a key pattern of one-to-many association.
B. Intermediate memory recollection. Input the key pattern to associate time series
patterns stored on the time A.
C. Recollection without hippocampus. Stop the output of hippocampus and amyg-
dala temporarily to investigate long-term memory formation on CX2.
D. Consolidate long-term memory with a long-term potentiation (LPT) process,
i.e., input the key pattern repeatedly. Hippocampus works to transform intermediate
memory to long-term memory.
E. Recollection of long-term memory. Stop hippocampus and amygdala to inves-
tigate long-term memory formation on CX2.
One simulation result of one-to-many association and long-term memory formation
are shown in Fig. 3. One time series pattern was stored in CX2, while the rate of stor-
age including failed one is reported by Table 3. The output of amygdala model during
storage and recollection processes is shown in Fig. 4. Comparing with conventional
hippocampus-neocortex model [8] [9], the model of limbic system proposed raised
8% rate of successful recollection.
