Adaptive Space Reconstruction and Hidden-level Generalization
Summary
Our living creatures represent global information in their brain
by integrating local sensory signals such as visual sensory signals.
In this paper, the state of hidden layer in layered neural network
with local inputs for some sets of training signals was observed.
Some characters became clear as follows.
(1)If the training signal changes gradually in space, the hidden layer
becomes to represent the spatial information.
(2)This tendency is stronger in the higher hidden layer.
(3)If there are redundant hidden neurons, they represent the global
information totally, while each of them keeps the initial fluctuation
due to the initial connection weights.
(4)If there is no correlation between the training signal of two input region,
the learning of one region becomes not to influence to the learning of the
other region.
(5)However, the hidden neurons does not become to represent the information
for only one region.
From these results, the reason why the hidden layer represents spatial
information by reinforcement learning and neural network\cite{DV}
can be thought as follows.
The state evaluation value changes gradually according to the time to the goal,
while motion should change gradually for the states whose evaluation values
are the same.
Keywords:
layered neural network, localized input, hidden representation,
generalization, learning
Reference
2. 柴田克成, 伊藤宏司:
局所信号を入力としたニューラルネットにおける中間層での適応的空間再構成と汎化,
電子情報通信学会技術報告, NC2001-152, pp. 151-158, 2002.3
(in Japanese)
pdf File (8 pages, 197kB)
1. K. Shibata and K. Ito :
Reconstruction of Visual Sensory Space on the Hidden Layer in a Layered Neural Networks,
Proc. of ICONIP (Int'l Conf. on Neural Information Processing) '98, Vol. 1, pp.405-408, 1998.10
[neural network, generalization, visual sensor, spatial recognition, integration]
pdf File (4 pages, 162kB)
Return to my home page (English)
Return to my home page (Japanese)