||Study on Properties of Periodic Chaos and Controlling Method of Islands in Chaos Neural Network Outputs
YOSHIDA, Hitoaki ,
SASAKI, Mitsuaki ,
MURAKAMI, TakeshiKAWAMURA, Satoshi
The 34th JSTE Tohoku Section Conference, Iwate University
60 , 2016-11-27
The properties of periodic chaos from CNN outputs changing the number of islands have been studied in detail. As the result of NIST SP800-22 tests the pseudo-random number from periodic chaos should not be used for security system. The correlation coefficient between subseries is normally negligible (<0.01), because the 2 subseries are independent. Periodic chaos, however, has only 2 modes, (i) two subseries simultaneously visit the same island (plus correlation), (ii) both series singly visit the other island (negative correlation). In this system only plus correlation is observed. Therefore periodic chaos can be numerically detected by correlation coefficient, so that periodic chaos is removed easily for a security system.