Action Strategies in Intelligent Reinforcement Learning Environments

• Main Authors: Pavel Basalin, Dmitrii Kulikov
• Format: Article
• Language: Russian
• Published: The Fund for Promotion of Internet media, IT education, human development «League Internet Media» 2021-04-01
• Series: Современные информационные технологии и IT-образование
• Subjects: hybrid intelligent learning environment; adaptation with reinforcement; strategies for choosing actions; primitive strategy; adaptive strategy
• Online Access: http://sitito.cs.msu.ru/index.php/SITITO/article/view/735

For a hybrid intelligent learning environment, strategies for choosing actions that are aimed at implementing events related to the study of fragments of the studied material are considered. One of them, the so-called primitive strategy, involves the implementation of each next initialized event from the most compact (least informative) action with transitions to actions of a more informative plan until the planned level of assimilation of the material is reached. In this case, when implementing each next event, the experience of implementing previous events is not taken into account, and the learning process may be unnecessarily cumbersome. Another strategy – adaptive – is aimed at using an action as a source for the implementation of the next event, an action that is an order of magnitude less capacious (compared to the final action of the previous event). A number of assumptions were formulated that allowed us to develop a set of rules that form the basis for the strategy of choosing actions in an intelligent learning environment (ILE). This algorithm, using feedback from the learner through the system's working memory (state matrix), implements a mechanism with reinforcement – the learning environment adjusts its behavior depending on the response of the external environment (the learner).