Gauss-Sigmoid Neural Network

Summary This paper introduces Gauss-Sigmoid neural network in which input signals are processed by gaussian units at first, and then the output signals are the input of Sigmoid-based neural network. It can approximate strong non-linear functions easily due to the local representation of gaussian units like Gaussian-based RBF (Radial Basis Function) network, and that can also represent global and abstract information like Sigmoid-based neural network by integrating many pieces of local information. The global and abstract information makes the learning of following related tasks drastically faster. In the hill-car task, in which Boyan et al. have shown that the combination of reinforcement learning and sigmoid-based neural network leads to instability of learning, the above advantages are examined.

Reference
7. 柴田克成, 前原伸一, 伊藤宏司:
Gauss-Sigmoidニューラルネット,
計測自動制御学会　システム・情報部門学術講演会2002 講演論文集, pp. 467--472, 2002.11.
(in Japanese)

6. 前原伸一，柴田克成, 杉坂政典:
強化学習によるGauss-Sigmoidニューラルネットの大域的情報表現の獲得,
平成13年度電気関係学会九州支部連合大会講演論文集, pp. 693, 2001. 10.
(in Japanese)

5. 前原伸一, 杉坂政典, 柴田克成:
連続値入力強化学習におけるGauss-Sigmoidニューラルネットの有効性,
電子情報通信学会技術研究報告, NC2000-166, pp. 75-82, 2001. 3.
(in Japanese)

4. 柴田克成, 前原伸一, 杉坂政典, 伊藤宏司:
Gauss-Sigmoid ニューラルネットワーク,
第12回自律分散システムシンポジウム資料, pp.133-138, 2001.1
pdf File (6 pages, 260kB)
(in Japanese)

3. 前原伸一, 杉坂政典, 柴田克成:
Gauss-Sigmoidニューラルネットワークを用いた強化学習の安定性,
第19回計測自動制御学会九州支部大会学術講演会予稿集, pp.475-478, 2000.11
(in Japanese)

2. Shin'ichi Maehara, Masanori Sugisaka and Katsunari Shibata:
Reinforcement Learning Using Gauss-Sigmoid Neural Network,
Proc. of AROB (Int'l Sympo. on Artificial Life and Robotics) 6th, Vol.2, pp.562-565, 2001.1
[reinforcement learning, Gauss-Sigmoid neural network, hill-car problem]

1. K. Shibata and K. Ito:
Gauss-Sigmoid Neural Network,
Proc . of IJCNN (Int'l Joint Conf. on Neural Networks) '99, #747 (6pages), 1999. 7
[neural network, localization, integration, RBF, learning]
pdf File (6 pages, 139kB)

Return to my home page (English)

Return to my home page (Japanese)