2019 ASEE Annual Conference & Exposition

The idea that a wide variety of domain-specific dynamics may be studied with a common set of mathematical tools can be traced back to Alexander Bogdanov's Tektology, Norbert Wiener's Cybernetics, and Ludwig Von Bertanlaffy's General Systems Theory.
In this NSF IUSE project, we propose to leverage recent theoretical advances in the field of {\it Coevolutionary Computation} to study, from a novel perspective, student vs. practice problems interactions. Using what we learned from bridging the above-mentioned fields, we also propose an algorithmic approach to generate pedagogically-sound practice problems.

In a seminal coevolutionary computation work by Hillis, a population of sorting networks was evolved against a population of sequences to be sorted.
The algorithm searched for sorting networks able to correctly sort sequences of numbers, while it also searched for sequences able to reveal flawed sorting networks.
Similar competitive interactions are also common in educational scenarios. For instance, a group of students might work on improving their skills with practice problems that are designed by educators to be increasingly challenging.

This similitude enables us to consider two novel research agendas.

First, we may leverage coevolutionary computation theories to provide a new perspective on educational research.
A large body of research in the field of coevolutionary computation has indeed been dedicated to investigating the reasons for which an ideal "arms race", in which both populations push each other to constantly improve, is actually difficult to obtain.
The so-called pathological coevolutionary dynamics that have been identified have direct counterparts in the educational domain.
It is therefore relevant to investigate the relevance of the solutions used to mitigate them in coevolutionary computation, and how they might apply to the educational domain.

Second, we may design new algorithms to coevolve practice problems against a population of learners.
We describe such an approach and discuss the roles played in it by various pathological coevolutionary dynamics, as well as how we mitigated them.

Even more interestingly, rather than only adapting the learning experience to each learner individually, such a system has the potential to identify misconceptions plaguing students at large, thus potentially assisting educators in engaging in Computing Education research on what makes a given topic difficult to learners.

We discuss the above points in the context of the specific system that has been developed in this NSF award.
Based on preliminary evaluations on both simulated benchmarks and actual students, we then make recommendations for the next step of a research agenda focused on what we termed {\it Coevolutionary-Aided Teaching} systems, and their potential to contribute to discipline-based educational research.