Informatics in Education

Debugging is integral to programming. It comes into play as soon as novices make their first mistakes in creating programming artifacts. It is also consistently reported to be a skill that is difficult to learn as well as to teach effectively. Research in Informatics Education has often focused on the process of debugging, by breaking it down in steps connected by temporal and causal dependencies. In this work, we focus instead on debugging as a skill, from the standpoint of Cognitive Load Theory, and break it down into a tree-shaped model of subskills that enable one another. Debugging may thus be seen as a meta-skill that requires the coordination of multiple others. From the standpoint of Cognitive Load Theory, such a skill is cognitively expensive, which may explain the learning-related difficulties tied to debugging. Using the framework of the four-component instructional design, we hypothesize a categorization of each debugging subskill as either recurrent or nonrecurrent, dividing those that are applied consistently to different contexts from those that require problem solving. All subskills may be practised and potentially assessed with targeted exercises, whose design depends on their recurrent/nonrecurrent nature. We provide extensive examples of such exercises. Our decomposition of debugging into subskills is a novel way to address debugging in educational contexts and complements the work done on debugging processes. Although it is currently a theoretically grounded conjecture, the model provides concrete guidance for instructors on analyzing existing materials and planning cognitive-load-informed learning trajectories.

The definition of effective pedagogical strategies for coaching and tutoring students according to their needs is one of the most important issues in Adaptive and Intelligent Educational Systems (AIES). The use of a Reinforcement Learning (RL) model allows the system to learn automatically how to teach to each student individually, only based on the acquired experience with other learners with similar characteristics, like a human tutor does. The application of this artificial intelligence technique, RL, avoids to define the teaching strategies by learning action policies that define what, when and how to teach. In this paper we study the performance of the RL model in a DataBase Design (DBD) AIES, where this performance is measured on number of students required to acquire efficient teaching strategies.