![](bg4.png)
m (=50) denotes the length of the sequence, and n
denotes the length of array (n=2 for pair). The more
the repeated pairs, the larger the value of RP,
indicate the deterioration of the memory capacity of
the subject. Since the case of n=3 did not show much
difference from that of n=2(pair), we stick to
consider only pairs (n=2). Note that RP ranges
[0:100] in percent, irrelevant to the size of the data
sequence unlike NSQ. We show in Figure 1 that the
data are separated to 4 distinct regions according to
the age groups by using RP, TPI, ADJ, H for indices.
Figure 1: The SOM representation of 20 subjects in RP,
TPI, ADJ, H. showing separation of different age groups:
A(20s), B(30-49), C(50-79), D(80-).
5 MOBILE PHONE KEYBOARD
HURG-on-MPK (Mobile Phone Keyboard) is
designed to reduce the length of data sequence,
which asks subjects to type 9 numerical keys on the
mobile phone keyboard once per each key in a
random order. In this scheme of HURG, the length
of data is fixed to 9, which is far shorter than the
previously studied HURG. Moreover, this is
effective to train the flexibility of brain, demanding
high level of concentration to the subjects.
This new method requires a new set of analytical
tools. Since all the 9 figures (1-9) are used in one
data only once, the randomness measure used for the
standard HURG such as entropy becomes useless in
this case. The randomness for HURG-on-MPK lies
in the order of those 9 figures.
We have developed a classification method of
such data by using a 3-layered feed-forward neural
network (3NN). The location the 9 figures plus the
total length of the path that the finger travels over
the keyboard are put into the 10 units of the first
(input) layer. Those are sent to the second (middle)
layer that consists of 3 nonlinear units, which
convert the weighted sum of the information from
the 10 input units into 1 (if it exceeds the threshold)
or 0 (if it is below the threshold). The outputs from
the 3 units of the middle layer are sent to the output
layer of the same kind of nonlinear structure and
they are compared with the teacher signals. We have
used the back-propagation learning algorithm for
training this 3NN. By using this, we have
successfully classified the 7 subjects. The rate of
recognition of 7 subjects (A-G) are shown in Table 3,
where the result with and without the 10-th unit are
compared. Note that the information of the total path
that the finger travelled put into the 10-th unit plays
an important roll.
Table 3: Recognition Rates [%] for 7 subjects (A-G).
Subject A B C D E F G ave
1-9 units 90 73 53 0 5
7
5
3
60 55
1-10 units 100 93 97 33 7
0
8
0
90 80
6 CONCLUSIONS AND BEYOND
We have presented in this article various ways of
pattern recognition of HURG, such as HMM,
correlation dimensions, etc., and the efforts to
shorten the length of data sequence. In this regard,
we discussed analytical techniques to extract
patterns from HURG, in particular, the identification
of the four indices, RP, TPI, ADJ, H to characterize
short sequences.
We have also introduced HURG-on-MPK and
presented the effectiveness of the 3 layered neural
network system (3NN), using the locations of 9
figures appeared in the data sequences and the path
length that the finger travels.
Our future work is to collect more data and test
the effect of HURG including the new method
proposed in this article. Other tools of pattern
recognition are to be considered.
REFERENCES
Wagenaar, W.A., 1977. Psychological Bulletin, 77, 65-72.
Towse, J., N., Nell, D., 1998. Analyzing Human Random
Generation Behaviour, Behavior Research Methods,
Instruments, & Computers, 30(A), 583-59.
Tanaka-Yamawaki M., 1998. Can We Measure the Brain
by Random Generations?, In Proc. ICONIP’98,
Omsha/IOS Press, 215-218.
Tanaka-Yamawaki M., 1999. Human Generated Random
Numbers and a Model of the Human Brain Functions;
Proc.1999 IEEE SMC, Piscataway, NJ, USA, 223-228.
Mishima, M., Tanaka-Yamawaki, M., 2008. Effective
Indices to Characterize Short Sequences of Human
Random Generations, Artificial Life and Robotics, 12,
Springer, 184-187.
HUMAN RANDOM GENERATION AND ITS APPLICATIONS
399