A study on integration of patterns and symbols for intelligent information processing

Control systems for actual plants with numerous variables have a difficulty in getting sufficient operating data that cover the whole input space. Another difficulty is to cope with defective inputs in the case of failure. These difficulties are unavoidable for posterior learning systems, such as artificial neural networks. One method to overcome these problems is to utilize humans´ knowledge. For easy utilization of humans´ knowledge, this study uses a visualized state map. Patterns in a multi-dimensional space are mapped into a 2-dimensional state map. If the patterns become visual, it is easy to be labeled. This is very important for the integration of patterns and symbols. This paper defines patterns and symbols. Two methods for conversion from a pattern to a symbol and its utilisation is proposed, and applied to simulations using an autonomous robot with distance sensors. The robot obtains patterns generated from inputs and its action is determined using symbolic processing