哔咔漫画

Improving Instruction with Coevolutionary Algorithms

By Brad Stager

Writing exams and creating exercises that can facilitate and assess student learning is a task that educators can sometimes find to be daunting. But research at the University of South Florida College of Engineering may yield tools that can make that part of the learning process easier to accomplish and pedagogically more productive.

The three-year, $377,012 National Science Foundation project, 鈥淯sing Coevolutionary Algorithms to Identify Distractor Answers for Multiple Choice Questions Used for Peer Instruction,鈥� builds on earlier research by Associate Professor Alessio Gaspar of USF鈥檚 Department of Computer Science and Engineering, who is the principal investigator.

鈥淭his new grant is a step toward further improving our techniques in order to evolve exercises that are usable in a much wider variety of STEM courses and topics,鈥� says Gaspar, who earned his PhD in computer science from the University of Nice Sophia-Antipolis in France.

鈥淒esigning a multiple choice question can take only a few minutes. Designing a good one in which each distractor option actually beacons to students featuring a specific common novice misconception, is much harder.鈥�

Gaspar is also director of the Computing Education Research & Evolutionary Algorithm Laboratory (CEREAL) group, which focuses on applying evolutionary algorithms and bio-inspired metaheuristics to advance computer assisted learning.

Improving students鈥� understanding of programming through solving puzzles was the research group鈥檚 initial application of evolutionary algorithms, an example of biomimicry that combines concepts of biological phenomena such as mutation and selection with the computational power of artificial intelligence. Programming exercises developed for students to work on were evaluated for relevance.

With the NSF grant, Gaspar and his team will be able to expand their research of evolutionary algorithms to develop software that will generate distractor (wrong) answers to multiple choice questions that optimally indicate student understanding. The wide use of multiple choice questions in education opens up possibilities of applying the research to other STEM fields and unrelated subject areas if the application to multiple choice questions proves to have merit.

鈥淢ultiple choice questions are usable in almost any course, sometimes as a form of summative assessment, but always as a great formative assessment tool,鈥� says Gaspar. 

鈥淗owever, their open-ended nature means that we cannot expect an algorithm to evolve them from scratch.鈥�

Creating materials that assess students鈥� learning has benefited from advances in learning technology such as using data mining, but there is room to improve efficiency and efficacy. According to Gaspar, the students themselves will contribute to the process of developing multiple choice questions.

鈥淥ur approach allows students themselves to cooperate to evolve distractors that are relevant to their own specific learning needs.鈥�

Gaspar adds that improving interaction between students, such as through peer instruction, is an important aspect of the research as well.

鈥淲e are going to use what we learned from these adaptive algorithms to improve students鈥� ability to collaborate with one another when learning new material.鈥�

Current tools such as intelligent tutoring systems focus on individual learners, whereas Gaspar鈥檚 research utilizes what a group of students may have in common and leverages that through algorithms.

鈥淎 population of learners co-adapts alongside a population of exercises. While the learners improve their skills as they practice, an algorithm evolves more relevant exercises.鈥�

Currently, Gaspar and his team are developing the first in a series of software tools that will be released over the course of the study and recruiting faculty interested in participating in the project. Besides developing new software tools for this specific project, Gaspar says there is an opportunity to explore new ways of developing and delivering knowledge through coevolutionary aided teaching.

鈥淭he underlying idea of having a population of learners co-adapt alongside a population of problems is a particularly exciting and novel approach to many educational problems,鈥� says Gaspar. 鈥淐oevolution can be a powerful nature-inspired metaphor to leverage in order to solve problems, albeit a difficult one to fully understand.鈥�