On neural networks for symbolic processing

摘要

Existing approaches to integrating neural and symbolic processing are divided into the following four categories: developing specialized, structured, localist networks for symbolic processing; performing symbolic processing in distributed neural networks (in a holistic way); combining separate symbolic and neural network modules; and using neural networks as basic elements in symbolic architectures (the embedded approach). Research issues that need to be addressed in order to advance this field as well as to better understand the nature of intelligence and cognition are outlined.