Title: Research on a computational model of learning to advance the theory of learning and teaching
Project Description: SimStudent is a computational model of learning. It is implemented as an interactive machine-learning agent with hybrid AI technologies involving inductive logic programming and heuristic search.
Using SimStudent, we study the theory of learning (how students learn) as well as the theory of advanced educational technology (how advanced technologies help students learn and researchers study how students learn).
There are currently three primary research streams as describes in SimStudent Projects: (1) teachable peer learner, (2) computational model of learning, and (3) intelligent authoring. Depending on your background and interests, you may find your own research topic on one of them (or their combination). You may also pioneer a new research agenda for the SimStudent project!
By working on the SimStudent project, you will learn (a) Java programming skills, (b) advanced statistical methods, (c) data mining techniques, (d) theories in the sciences of learning, and/or (e) HCI methods for a usability study.
Here is a list of some potential projects (you should be creative!):
(1) Data-driven Model Improvement Study – Can SimStudent make a better cognitive model for a cognitive tutor? The effectiveness of a cognitive tutor largely depends on the quality of a cognitive model, which represents "skills" to solve problems that students must learn. Domain experts and researchers often spends hundreds of hours to design and implement a cognitive model. The cognitive model in this context is a set of production rules that is exactly what SimStudent generates. In theory, therefore, we should be able to use SimStudent to create high-quality cognitive model by simply tutoring SimStudent. Primary goal of this project would to test if this simple (but fascinating) idea works by actually implementing an example cognitive tutor and apply SimStudent to it. No particular research experience might be required, but basic programming skills in Java would a plus.
(2) Educational Data-mining Study – How do students learn by teaching a synthetic peer? Using the actual study data we have collected in classroom studies, a primary goal of this project would be to explore cognitive and social mechanism that mediated tutor learning (see details about our research on learning by teaching). Knowledge and experience in advanced statistical analysis and/or data-mining techniques would be required.
(3) Usability Study – What makes the Learning-by-Teaching environment more user-friendly, hence facilitating better tutor learning? From a human-computer interaction point of view, improvement of the systems’ usability is an essential key for success in accomplishing our research agenda. In this line of research project, the REU student intern would apply various HCI methods to evaluate the system’s usability and explore key HCI factors to maximize tutor learning. Knowledge and experience in HCI methods would be required.
(4) Prior Knowledge Study – How does the “individual” differences of SimStudent (i.e., the tutee) affect the student’s (i.e., the tutor’s) learning? By manipulating the background knowledge of SimStudent, we can control the speed and accuracy of SimStudent’s learning. For example, SimStudent may start with a certain amount of knowledge for equation solving, or SimStudent might have immature or even irrelevant background knowledge that slows down learning rate and causes more errors. The goal of an REU project would then to study how such differences affect the tutor learning.
(5) Pedagogical Agent Study – How would the appearance of SimStudent and its functionality affect the tutor learning? What if SimStudent has emotion and expresses its affective status? Can SimStudent share its affects with the student and if so how would such a sympathetic pedagogical-agent influence the tutor learning? The goal of an REU project would be to study an affective interaction between SimStudent and human student and to understand how such emotional interaction would affect tutor learning.
(6) Machine-Learning Study – Can we improve the SimStudent’s learning algorithm? So far, we used inductive logic programming that is basically implemented as a brute forth search. We also use FOIL (Quinlan, 1990), which learns Horn Clauses from relations provided in examples. There are pros and cons for the current implementation (mostly the issues for domain generality and knowledge representation). The REU on this project would study alternative technologies to enhance the generality and efficiency of SimStudent’s learning.
Noboru Matsuda <Noboru.Matsuda@cs.cmu.edu>
Human Computer Interaction Institute
Carnegie Mellon University
5000 Forbes Ave. Pittsburgh, PA 15213
Voice: 412-268-2357 Fax: 412-268-9433