Intelligent Learning Material Delivery Agent (ILMDA)
Research
There are two groups of research areas. In terms of fundamental CS research, we have focused on intelligent agents and machine learning. In terms of intelligent agents, we look at multiple fault diagnosis and agent awareness. LD Miller and Akira Endo’s research will address these two topics. In terms of machine learning, we look at case-based learning (adaptation heuristics, similarity, fuzzy database retrieval, case storage, and reinforcement learning). Todd Blank’s research has focused on adaptation heuristics, relaxed database retrieval and case storage. Ashok Thirunavukkaras addressed part of the relaxed database retrieval problem.
Development
In terms of development, our ILMDA tutoring system is currently deployed at the CSCE155 lab, with rudimentary GUI, and prototype modules. The case-based reasoning infrastructure is stable and operational. The learning component is still in its infancy as agent diagnostics and awareness is being formalized at this moment. The relaxed database retrieval has been prototyped but the fuzzy relaxation still needs significant improvement. Currently, our graph traversal and simulated annealing components, though operational, have been taken out of the ILMDA operational system. We will consider incorporating these two components into future versions.
In terms of the support software, we have built two authoring interfaces, one for content development (we used this interface to author our learning objects: tutorial, examples, and problems), and one for results reporting (this has recently been completed and will be deployed in Fall 2005). We have also built a primitive simulator that we used to test our machine learning component.
In terms of content development, we have built five learning objects for CS1 topics: File I/O, Exceptions, Event-Driven Programming, Inheritance and Polymorphism, and Recursion. Each learning object has a tutorial, 4-5 examples, and 20-25 problems.
Experiments
Preliminary results were collected from Fall 2004 deployment. Results were encouraging in terms of software use; in terms of student learning, they were inconclusive. Both Todd Blank and LD Miller are working on finalizing their analysis of the data.
Based on Fall 2004 deployment results, experiments were modified for Spring 2005 to better analyze the machine learning impact. Todd Blank is assimilating the results.