Behavioural cloning of teachers for automatic homework selection
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
MetadataShow full item record
Moore, R., Caines, A., Rice, A., & Buttery, P. (2019). Behavioural cloning of teachers for automatic homework selection. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 11625 LNAI 333-344. https://doi.org/10.1007/978-3-030-23204-7_28
© Springer Nature Switzerland AG 2019. We describe a machine-learning system for supporting teachers through the selection of homework assignments. Our system uses behavioural cloning of teacher activity to generate personalised homework assignments for students. Classroom use is then supported through additional mechanisms to combine these predictions into group assignments. We train and evaluate our system against 50,065 homework assignments collected over two years by the Isaac Physics platform. We use baseline policies incorporating expert curriculum knowledge for evaluation and find that our technique improves on the strongest baseline policy by 18.5% in Year 1 and by 13.3% in Year 2.
External DOI: https://doi.org/10.1007/978-3-030-23204-7_28
This record's URL: https://www.repository.cam.ac.uk/handle/1810/300988