A study of building intelligent systems using reinforcement learning for remote instrumentation and immersive learning
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
Reinforcement Learning (RL) has the potential for developing intelligent, adaptive systems and foster feedback-oriented training of software to make decisions that maximize rewards, and increase accuracy of outcomes. However, challenges such as the complexity of real-time analysis in dynamic environments, ensuring data quality and diversity for robust training, and making RL models interpretable for stakeholders present unique research challenges. In this thesis, we explore the design and implementation of RL-based algorithms to develop intelligent systems within two distinct domains: (i) remote instrumentation for intelligent image analytics in materials manufacturing, and (ii) virtual reality for intelligent pedagogical assistance in immersive cybersecurity education. In the first domain, we present the design of an intelligent system for real-time image processing tasks in a cloud-based platform designed to automate and optimize the process of image analytics in scientific experiments. Specifically, we develop a Remote Instrumentation Science Environment (RISE) platform that integrates remote instruments and data centers using cloudlets, providing seamless access through a high-level language API. By employing RL agents, RISE enables real-time analysis and dynamic adjustment of instrument settings enhancing experimental control. In the second domain, we present the design of an intelligent Pedagogical Agent (PA) to support adaptive learning experiences in a Virtual Reality Learning Environment (VRLE), "CyEscape". Integrating Deep Q-Network with OpenAI and Unity, the PA is trained on guiding the learners through a dynamic learning environment, featuring escape rooms that create a practical and immersive approach to cyber security education. Through validation experiments of RISE and CyEscape, we highlight the transformative potential of RL in creating intelligent systems that are adaptive, and user-centric to improve pace of discovery in image analytics, and educational outcomes in cybersecurity training.
Table of Contents
DOI
PubMed ID
Degree
M.S.