Pr. Cristina Conati / University of British Columbia
There is extensive evidence that AI-based educational technology can effectively provide personalized support to help students learn problem solving skills in a variety of domains. In contrast, until recently there has been limited work on AI-based environments to support educational activities that are more exploratory in nature, such as learning from interactive simulations or playing educational games. These activities are becoming increasingly widespread, especially in the context of distant learning and other forms of self-directed learning, because they can increase motivation and foster grounded skills acquisition. However, not everyone can learn effectively from these activities, calling for AI-driven Learning Environments that can provide personalized support for open-ended exploratory learning, while interfering as little as possible with the unconstrained nature of the interaction. In this talk I will discuss the unique challenges of this endeavor, and our proposed solutions to address them, including how to devise AI-driven models that can track and react to open-ended behaviors beyond those traditionally addressed by analogous models for problem solving. I will also present our most recent results on the potential benefits of making such models explainable to their end users, especially if the explanations are personalized to each user’s specific needs.
Pr. Sidney D'Mello / University of Colorado Boulder
Educational data mining (EDM) has made tremendous progress in using click-stream log-data to model, adapt, and personalize to individual students. But learning is inherently a social endeavor involving interactions among living and breathing people, not cold, disembodied human-machine transactions. Thus, modeling and supporting groups of students in collaborative learning is an exciting frontier for EDM, especially for naturalistic interactions involving speech and language, posture and gesture, facial and head movements, and central and peripheral physiology. I’ll discuss emerging research on multimodal, multiparty modeling and facilitation of collaborations among students, couples, and teams in the lab, the classroom, and workplace. I’ll also discuss the NSF National Institute for Student-Agent Teaming which aims to: develop the theories, Artificial Intelligence (AI) technologies, and know-how for creating next-generation collaborative learning environments composed of diverse students, teachers, and AI.
Pr. Pierre Dillenbourg / École Polytechnique Fédérale de Lausanne
Entering a modern car is like entering a computer with wheels, seats and windows. Similarly, entering a classroom is like entering a large digital system with chairs, windows and a board. The input devices of this system are not a keyboard and mouse, but an entire classroom and its various sensors. The output device of this system is not a screen but a set of digital elements distributed in the class. The output is not a simple reflection of the input but input data are processed by multiple operators that aggregate, compare and visualize data. The resulting dashboards are used for monitoring the learners’ progress in order to decide when and to whom to intervene. They are also used to compile data from the constructivist activities for supporting the debriefing phase, as well as to predict the completion time of an activity. Monitoring, debriefing and timing are central processes in classroom orchestration.
Pr. Cristóbal Romero / University of Córdoba
This Talk is about predicting or classifying student’s performance starting from student’s usage/interaction data with learning environments. This is one of the most important tasks in Educational Data Mining (EDM) and Learning Analytics (LA) research communities. The first part of the talk describes the original paper presented in EDM’08 in Montreal as full paper. It compares different data mining algorithms provided by Keel DM software for classifying students based on both students’ usage data in 7 Moodle courses and the final marks obtained in the corresponding Cordoba University exams. The second part of the talk describes new research lines and improvements from 2008 to the present (2021): other DM tools/software and frameworks, other classification methods/algorithms, meta-learning for parameter tuning/algorithm selection, other evaluation metrics and statistical tests, the Baker LAP criteria, early warning prediction and more data from Multisource, Multimodal and Smart learning.