Modeling Human Learning Characteristics for Analyzing Human Teaching Methods

This study aims to clarify the characteristics of human learning processes and to examine the requirements for a trainee model that can reproduce them. The results of a Human– Human teaching experiment suggest that human learning processes consist of two phases: an initial phase where performance improves significantly with large variability, and a later phase where performance improvements slow down and variability converges. However, the proposed trainee model exhibited issues such as instability in early learning and performance degradation, despite being able to reproduce the learning progress. To address these problems, a learning rate scheduler that dynamically adjusts the learning rate according to the teaching force was introduced. This suppressed the instability in early learning and mitigated the performance degradation. These results suggest that the characteristics of human learning processes can be reproduced to some extent. In the future, it will be necessary to introduce probabilistic mechanisms such as Bayesian networks and develop dynamic parameter determination methods.