Yukio Hayashi
Oscillatory Neural Network and
Learning of Continuously Transformed Patterns
Abstract:We investigate experimentally the dynamic behaviors of an oscillatory neural
network. Computer simulations show an interesting characteristic: the autonomous
generation of a limit cycle near a memory (memory retrieval with ambiguous
fluctuation) for an input near a memory, and of a chaotic orbit among
memories (autonomous search) for an input far from memories. We also analyze
theoretically a few restricted behaviors near a memory.
This type of neural network can treat spatiotemporal pattern processing in the
brain. As an example of dynamic information processing, it is shown that continuously
transformed pattern cycles for three Japanease characters are embedded
in the limit cycles of the oscillatory neural network by a learning method. The
existence in the brain of a continuously transformed. pattern operation for a character
is also discussed from the cognitive psychological point of view.
The characteristic behavior of limit cycle or chaos according to an input in our
oscillatory neural network may be useful for developing a dynamic information
processing mechanism for a spatiotemporal pattern in the brain.
Keywords: Oscillatory neural network, Dynamic information processing, Spatiotemporal pattern,
Limit cycle near a memory, Chaotic orbit among memories, Continuous transformation, Learning
of recurrent network