Presentation 2019-10-12
Stability of Periodic Orbits in 3-Layer Dynamic Binary Neural Networks
Seitaro Koyama, Toshimichi Saito,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The 3-layer dynamic binary neural networks (3-DBNNs) are characterized by ternary connection parameters and signum activation function. Depending on the connection parameters, the 3-DBNN can generate various binary periodic orbits. The 3-DBNN is based on the 2-layer DBNN discussed in our previous publications. Applications include associative memories and central pattern generators. In this paper, we analyze the dynamics of the 3-DBNN that generates a target binary periodic orbit. In order to analyze the dynamics, we introduce two feature quantities. The first quantity characterizes stability of the target periodic orbit. The second quantity characterizes sparsity of the connection. We consider relationship between stability of the target periodic orbit and sparsity of the connection. We can see that as the connection sparsity increases, the orbit stability tends to be strong. As the connection sparsity increases further, the orbit stability becomes weak.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Dynamic binary neural networks / Associative memories / Periodic orbits / Stability
Paper # MBE2019-39,NC2019-30
Date of Issue 2019-10-04 (MBE, NC)

Conference Information
Committee MBE / NC
Conference Date 2019/10/11(2days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Taishin Nomura(Osaka Univ.) / Hayaru Shouno(UEC)
Vice Chair Takashi Watanabe(Tohoku Univ.) / Kazuyuki Samejima(Tamagawa Univ)
Secretary Takashi Watanabe(Kyushu Univ.) / Kazuyuki Samejima(NAIST)
Assistant Yasuyuki Suzuki(Osaka Univ.) / Akihiro Karashima(Tohoku Inst. of Tech.) / Takashi Shinozaki(NICT) / Ken Takiyama(TUAT)

Paper Information
Registration To Technical Committee on ME and Bio Cybernetics / Technical Committee on Neurocomputing
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Stability of Periodic Orbits in 3-Layer Dynamic Binary Neural Networks
Sub Title (in English)
Keyword(1) Dynamic binary neural networks
Keyword(2) Associative memories
Keyword(3) Periodic orbits
Keyword(4) Stability
1st Author's Name Seitaro Koyama
1st Author's Affiliation Hosei University(HU)
2nd Author's Name Toshimichi Saito
2nd Author's Affiliation Hosei University(HU)
Date 2019-10-12
Paper # MBE2019-39,NC2019-30
Volume (vol) vol.119
Number (no) MBE-224,NC-225
Page pp.pp.51-51(MBE), pp.51-51(NC),
#Pages 1
Date of Issue 2019-10-04 (MBE, NC)