Search for: All records

Public Large Language Models (LLMs) are being increasingly explored for aiding students to learn difficult concepts in STEM education. However, public LLM-based chatbots do not provide course-regulated answers that are important in a tutoring context, where instructors do not want students to work with chatbots on out-of-scope material, seek answers that hinder learning, or to bypass curricular sequencing. In this paper, we present “SAGE” (Student-focused Adaptive Guidance Engine) as part of an intelligent tutoring framework that can enforce instructor-regulated learning through goal-oriented and learning state based content access controls to ensure scaffolded, curricular sequencing and stepby- step guidance of students. SAGE access control features a three-layer gatekeeper mechanism that combines role-based, attribute-based, and cognitive state–based controls built on top of a knowledge graph of learner state and related course content. We evaluate SAGE in an exemplar programming course scenario and compare its performance against state-ofthe- art LLM baselines using a custom educational benchmark. Experimental results on the benchmark show that promptbased tutors (42%) perform similar to SAGE (44%) on metrics of helpfulness owing to prompt engineering. However, they frequently violate instructor policy boundaries, with SAGE providing 90% security for curriculum access control policies compared to 30% for prompt-based tutors. 

Dialog systems (e.g., chatbots) have been widely studied, yet related research that leverages artificial intelligence (AI) and natural language processing (NLP) is constantly evolving. These systems have typically been developed to interact with humans in the form of speech, visual, or text conversation. As humans continue to adopt dialog systems for various objectives, there is a need to involve humans in every facet of the dialog development life cycle for synergistic augmentation of both the humans and the dialog system actors in real-world settings. We provide a holistic literature survey on the recent advancements inhuman-centered dialog systems(HCDS). Specifically, we provide background context surrounding the recent advancements in machine learning-based dialog systems and human-centered AI. We then bridge the gap between the two AI sub-fields and organize the research works on HCDS under three major categories (i.e., Human-Chatbot Collaboration, Human-Chatbot Alignment, Human-Centered Chatbot Design & Governance). In addition, we discuss the applicability and accessibility of the HCDS implementations through benchmark datasets, application scenarios, and downstream NLP tasks. 

As Virtual Reality (VR) gains traction in education, its potential to support neurodivergent learners in cybersecurity training remains underexplored. This emerging technology report examines how VR can bridge the gap between STEM education and cybersecurity training for neurodivergent individuals, highlighting both its promise and the challenges that must be addressed. While VR-based cybersecurity simulations offer immersive, hands-on learning experiences that align with neurodivergent strengths, existing implementations often overlook critical accessibility considerations. This emerging technology report reviews current VR-based cybersecurity training systems, and provides insights and limitations in how they are supporting neurodivergent users. This report also addresses key challenges in this area of research such as cybersickness, the lack of neurodivergent representation in VR development, and the difficulty in creating realistic cybersecurity simulations. Given the rapid evolution of VR in cybersecurity education, ensuring accessibility requires intentional design choices and co-development with neurodivergent learners. We conclude by identifying research gaps and advocating for a more inclusive approach to VR-based cybersecurity education that fosters diversity within the field. 

Edge intelligence enables distributed decision-making by executing complex optimization tasks directly on resource-constrained edge nodes, minimizing reliance on centralized cloud infrastructure. This work introduces a hybrid edge-intelligent routing framework that combines Graph NeuralNetwork (GNN)–based graph pruning with metaheuristic optimization to achieve scalable and low-latency routing at the network edge. The GNN functions as a lightweight inference module that identifies and removes low-utility edges from a sensing network, substantially reducing communication and computational overhead. On the resulting sparse subgraph, a Guided Local Search (GLS) algorithm performs localized route refinement to produce near-optimal paths without central coordination. This integrated GNN–GLS design al-lows simultaneous graph learning and optimization on edge hardware while retaining solution quality comparable to cen-tralized solvers. Experimental results on synthetic datasets demonstrate a 13.2% runtime improvement on 1000-node graphs with only a 0.7% increase in route length, confirming the feasibility of learning-driven pruning and decentralized metaheuristic search for scalable edge deployment. 

The uSucceed project aims to support neurodiverse individuals in the STEM workforce by utilizing Virtual Reality (VR) to deliver a customized training curriculum in CyberSecurity. This short paper delves into the design and methodology implemented by the uSucceed learning system. Preliminary usability test evaluations by neurodiverse individuals (n = 8) reveal critical insights into user experience, particularly regarding cybersickness and the usability of the uSucceed VR learning system. Usability findings revealed positive feedback on the immersive environment but highlighted issues with task navigation and inconsistent responses from the AI-driven pedagogical agent. Cybersickness levels ranged from low to moderate, with dizziness and eyestrain being the most reported symptoms. These results serve as a framework for further refining of the curriculum and system design to enhance usability. As the project evolves, it is moving towards the enhancement phase of the learning system’s development, with a focus on further advancement of the context-driven AI pedagogical agent.