salle 206
17 May 2018 - 14h00
Automatic Grading: take a CEGAR and let the machine do your work
by Michael PERIN from VERIMAG / Université Grenoble-Alpes
Abstract: Working as an assistant professor is great, but with one dark side: grading exams is boring and takes time. The unavoidable consequence is non-uniform marking due to the weariness of a human corrector. Moreover, it provides few and delayed feedbacks to students.
Nowadays almost all students work on their own laptops and it thus becomes feasible to use these for digital exams. For teaching units where the correctness of an answer is decidable, digitalization could potentially lead to (1) the teachers' dream of automatic grading and (2) the possibility for students to train with high frequency feedback through automatic correction.
The problem of grading exams is not detecting the perfect answers but (3) finding good ideas that are worth some points in incorrect but partially correct answers (95% of answers).
I will present a Counter-Example-Guided Answer Repair (CEGAR) method addressing the problems (1,2,3) and illustrate it on the case of finite word automata.
This proposal is inspired from the CEGAR verification technique in software validation where CEGAR stands for Counter-Example-Guided Abstraction Refinement.
