Amruth N. Kumar
Our work on courseware involves generating problems to test a user’s learning. We propose to generate these problems dynamically in order to present an endless supply of problems to the user. In order to facilitate the user’s learning, we want to vary the challenge of the dynamically generated problems. Our current task is to identify a measure of challenge in problems on a topic, so that we can systematize and automate variations in the challenge of dynamically generated problems.
We propose to use a measure of challenge of problems based on analysis of the knowledge/ability necessary to solve problems in a given domain. By limiting such knowledge/ability to the minimum necessary to solve a problem, we attempt to avoid counting as evidence of learning, any serendipitous solving of problems.
In this paper, we analyze two different problem domains in Operating Systems in an attempt to derive measures of challenge for problems in those domains: storage placement, and processor scheduling. For these domains, we identify taxonomies of problems based on the minimum knowledge/ability required to solve them. We tentatively generalize the taxonomies to derive a measure of challenge for a class of scheduling problems dealing with discrete quantities.