Calibrating Educational Design Quality by Integrating Theory and Practice Using Neural Auto-encoders Home : Annual Conference 2017

Public Group Active 3 years, 7 months ago

Thu, Sep 7 2017, 1:00pm – 2:40pm

Authors: Alaa Bafail, Jonathan Tepper, and Ann Liggett

Room: Mountford Hall (G)

Theme: At the forefront of innovation

Type: 20-minute session

Description

Current learning design systems, such LAMS (Dalziel, 2003) or LDSE (Lauriallad et al’s, 2011), do not either encode a quantitative measure on which to base alignment or any means in which to adapt design patterns according to the student experience. This paper therefore introduces a neurally-inspired approach to learning design that seeks to address this gap by implementing Tepper’s quantitative measure of constructive alignment (Tepper, 2006) and associating module design patterns with their student satisfaction levels to calibrate alignment measures. The model, we call the Educational Design Intelligence Tool (EDIT), consists of a neural auto-encoder which is trained on 519 design patterns spanning 476 modules from the STEM discipline. Neural auto-encoders learn to output the module design pattern that is on its input layer and thus form a perfect memory of the training patterns. The intention here is that if trained with only design patterns eliciting high levels of student satisfaction it will form an internal memory of ‘good’ design patterns and therefore during testing, when presented with design patterns having low student satisfaction, it will attempt to associate it with one or more of the nearest matching ‘good’ design patterns it has encoded. We infer this as the model making suggestions to ‘enhance’ the module design and thus attract higher level of student satisfaction. Furthermore, we apply statistical analysis and self-organising maps to visualise and evaluate the changes recommended by the neural auto-encoder to identify the underlying ‘design preferences’ it has discovered from the data.

EDIT, with its data-orientated and adaptive approach to design, reveals orthodox practices whilst revealing some unexpected incongruity between alignment theory and design practice. For example, as expected, increasing the amount of questioning, interaction and group-based activity effects higher levels of student satisfaction even though misalignment may be present. However, the model is relatively ambivalent towards the alignment of learning outcomes and learning objectives suggesting there is some confusion between practitioners as to how these are related. The presentation aims to present a model that has been developed to actually modify the module design to promote constructive alignment and higher student satisfaction. This includes examples of modifications made by the system for a number of science-based modules. Subsequently, the audience will be made aware of a tool, based on deep learning that is able to enhance their module designs.

KEYWORDS: Learning design, Constructive alignment, Bloom’s Taxonomy, Auto-encoder networks, Artificial Neural Network

References:

Dalziel, J. (2003), Implementing Learning Design: The Learning Activity Management System (LAMS). Proceedings of the 20th Annual Conference of the Australasian Society for Computers in Learning in Tertiary Education.

Laurillard, D., Charlton, P., Craft, B., Dimakopoulos, D., Ljubojevic, D., Magoulas, G., Masterman, E., Pujadas, R., Whitley, E.A. and Whittlestone, K. (2011), A constructionist learning environment for teachers to model learning designs. Journal of Computer Assisted Learning, 29: 15–30.

Tepper, J. A (2006), Measuring constructive alignment: an alignment metric to guide good practice,” in 1st UK Workshop on Constructive Alignment, Higher Education Academy Information and Computer Sciences (ICS), Subject Centre and Nottingham Trent University, UK.

Go Back