Memory-based neural network and its application to a mobile robot with evolutionary and experience learning
Use of the neural network in pattern recognition problem has many beneficial aspects, including advantages in learning, generalization, and robustness. However, the use of neural networks also has drawbacks. Problems encountered during the use of neural networks include an extended learning period and the inability of the users to process data. In an attempt to overcome these disadvantages, we propose a memory-based implementation of neural networks. Our method realizes neural network-like properties such as learning, generalization and robustness and is free from the weak points of the neural network. In our approach, training data are stored in a memory in the form of distributed manner by the use of several random number tables. On-line learning can be realized easily in our approach. This method was applied to a behavior-learning mobile robot. This robot acquires instinctive behavior by evolutionary method.
Unable to display preview. Download preview PDF.
- 1.P. Kanerva, Sparse Distributed Memory, MIT Press, 1988.Google Scholar
- 2.T. Furuya, Probabilistic Distributed Memory, IJCNN, 1992, BeijingGoogle Scholar
- 3.T. Furuya, Self-Programming Network(SPN), IJCNN, 1993, NagoyaGoogle Scholar
- 4.R.A. Brooks, New Approaches to Robotics, Science, vol. 253, 1991.Google Scholar
- 5.H. Ito, Intelligent mobile robot, ICNN, 1995, PerthGoogle Scholar