Introduction and Motivation

Abstract

Flexible information processing, in which intuitive information can be processed in a human-like manner, is believed to be the next long-term goal in the evolution of the information society for the next century [1]. Computers and robots are expected to be able to perform their tasks in a real-world environment full of uncertain and ambiguous information. This is in contrast to the present situation where conventional “hard” information processing by computers is based on the assumption of completely-given information in a pre-assumed world or problem domain. Bringing computers from the comfort of specialized laboratories into the real world, where tremendous amounts of visual, speech, and other information of many kinds surrounds them, presents a challenge of a much different kind. Besides requiring a much greater computational power, there remains a considerable theoretical challenge to solve many problems barely comprehended or yet not even identified. Ideas such that information-storage and processing modules should be integrated together, that memories and databases should be self-organizing, that computers should understand conversational speaker-independent speech, are merely hints at a journey toward future adaptable autonomous systems [1]. Clearly, now, we are far from the time that “We simply explain what we want, and then let our machines do experiments, or read some books, or go to school —the sort of things that people do— [2]” to fmd a solution. However, scientists have begun to dream technically about such possibilities. Amongst strong evidence of the future potential, are simple learning machines currently available [3, 4].