Improving Cognitive Skills with a Multi-Agent DQN: A MATLAB Implementation
A. Dimitrova, P. Georgieva · 2025
This paper presents Multiple Intelligences Agent Deep Q-Network (MIADQN) which is a novel MATLAB-based framework for evaluating and enhancing cognitive skills in humans using a deep reinforcement learning. MIADQN integrates Multi-Agent Deep Q-Networks and the Theory of Multiple intelligences, developed by Howard Gardner. In the system each form of intelligence is modeled as an autonomous agent capable of real-time interaction with other agents and a dynamic, adaptive environment. The MIADQN architecture promotes interagent communication and decision-making, focusing on personalized learning experiences and advancing cognitive growth. Applying the MATLAB computational capabilities, the framework integrates deep reinforcement learning methodologies and enables effective management of shared memory across agents. Preliminary test results demonstrate the framework's ability to identify dominant intelligence and dynamically adjust to users' progress. With future development of the system, there is a potential to support people's individualized instruction across diverse educational settings, including online and hybrid learning environments. By emphasizing adaptability and diversity, this paradigm lays a foundation for future innovations in kindergartens, as well as in STEM and distance learning in schools and universities, integrating artificial intelligence techniques with principles from educational psychology to redefine cognitive skill development.