KNOWLEDGE ACQUISITION AND LEARNING FOR EXPERT SYSTEMS: RESEARCH ACTIVITIES IN JAPAN

Abstract

This chapter describes knowledge acquisition and learning for expert systems. Most of the existing knowledge acquisition systems are based on top-down elicitation, which is an interview for obtaining task-specific knowledge conducted assuming that the task type is known in advance. Knowledge compilation is a very effective method for acquiring heuristics from a model describing a working mechanism for a system where deep functional knowledge is available. The need for this approach has been recognized since the early phase of expert system technology, and significant work has already been carried out on compiling diagnostic knowledge from deep models. Deep knowledge is objective knowledge that depends on domain, not on task. Task knowledge is embedded in the compiler. The underlying concept is that many useful heuristics can be justified by deep objective knowledge. It would be desirable if deep knowledge could be prepared in a task-independent manner and if various task-dependent knowledge compilers worked on this knowledge to produce task-specific shallow knowledge.