William Cope
Adam P Rusch
Jason Berg
Duane Searsmith.png "Rita van Haren")
Rita van Haren
Eric Patnoudes
Maura Rodgers Brand
Jason Williams
Melanie Lamus
Heather A Ochmane-Learning Ecologies MOOC’s Updates
Khan Academy as a Case Study in Adaptive Learning
One topic that was briefly touched on in the lectures, which happens to be an area that I'm particularly interested in, is the idea of adaptive learning. The simplest example of this, which was discussed in the lecture videos, is a test or task that adapts to the level of the learner. The GRE used this technique for several years: within each section, each successive question would get easier or harder depending on whether the previous question was answered correctly.
However, this is only the beginning of the possibilities for offering adaptive, differentiated learning for students of different levels and backgrounds. At Khan Academy, a non-profit that was among the earliest movers in the effort to provide online education to a massive audience, the question the system asks is much more complicated: at what point has the student achieved mastery of the material?
Their solution involves machine learning (based on an explanatory blog post by a Khan Academy intern, which can be read here). To implement it, they needed a quantitative definition of mastery. In this case, they chose a probability threshold for getting the next question right. That is, based on your performance on the previous questions in the set, what is your chance of getting the following question correct? If it is above a certain percentage, the system considers you to have mastered the material.
With this in hand, the Khan team was able to use a regression model to calculate student mastery. They combine this with ongoing feedback (in the form of badges, 'hot streak' notifications, etc.) to let the students know how they are doing.
This is a fairly sophisticated technique by the standards of today, but in theory machine learning could be used to measure student performance across a much wider set of dimensions. A more advanced future system could provide students with question types and input types (e.g. video, audio, text) that most suit their learning process.
Sources:
http://magoosh.com/gre/2012/is-the-revised-gre-adaptive/
http://david-hu.com/2011/11/02/how-khan-academy-is-using-machine-learning-to-assess-student-mastery.html
http://www.khanacademy.com
Thank you Jared for introducing this topic. Adaptive tests are extremely important for motivation of students. When questions are too difficult or too easy students are discouraged. Even the gamers are discouraged when missions are too difficult to complete. I think quizzlet.com cards and tests use this principle too.
Thanks Jared for the excellent explanation of how adaptive learning works in Khan Academy. Without technology it would be hard to implement differentiated learning in any educational context. However, advances in recent technologies applications and tools allows us make things easier and find ways to customize learning for every learner.