Abstract
We propose an information-media theory of learning, and to incorporate the learning mechanism of human beings into machine learning. In the theory, a learning is regarded as communication by using multiple information-media, each of which includes a specific past experience forming imagery. This paper reports an experimental investigation of human learning processes dynamically selecting information-media, and raises a research issue on learning mechanisms dealing with multiple information-media.
The authors perform psychological experiments, in which subjects learn the programming language Prolog, and are asked to solve some questions. The subjects are divided into three groups according to their textbooks: (A) one containing only the specification of Prolog language, (B) the specification and an information-media explained in sentences, and (C) those explained in sentences and diagrams. Their incorrect answers, protocols, and elapsed time for solving questions are analyzed to clarify timing introducing a new information-medium, their selected media, and the effect of diagrams.
Based on the experiment, the authors constructed a cognitive model for simulating the behavior of subjects. The model is programmed in Prosit, by which each medium is handled as a situation. It confirms the information-media theory, and is a prototype towards a knowledge-intensive and robust machine learning mechanism.
