The Flipside of Machine Learning

Image: ShutterstockHuman learning is tricky and incredibly individual. Retaining the knowledge to pass a physics test may come quickly and easily to one student, but may require hours of cramming and consideration for another.

Professors at the Univ. of Wisconsin-Madison are melding the fields of computer science and psychology to reverse engineer machine learning, with the hope of devising ideal lesson plans to ease the learning process.

Called “machine teaching,” the method inverses the idea of machine learning.

Machine learning “is a mathematical procedure,” explains university computer scientist Jerry Zhu in an interview with R&D Magazine. “The mathematical procedure takes in training data (i.e. lesson) and finds a function that fits the training data. Importantly, the function can ‘generalize’ (i.e. extrapolate) beyond the training data—this gives the machine learning the ability to answer new questions.”

Machine teaching works backwards, as it already knows what knowledge needs to be imparted on the student. The concept “uses sophisticated mathematics to allow researchers to model actual human students and devise the best possible lessons for teaching them. While the definition of ‘best’ in a particular setting is up to the teacher, one example could be identifying the smallest number of exercises for a particular students to grasp a concept,” according to the Univ. of Wisconsin.

Funded by a two-year seed grant from the university, university computer scientist Jerry Zhu and Timothy T. Rogers, a professor of cognitive psychology, are exploring the concept and hope to make machine teaching a reality.

“The seed grant started this summer,” says Zhu. “First of all, machine teaching is still basic research, heavy on math and theory. Don’t expect it to show up in classrooms next month … Right now, we are taking baby steps to apply the methodology to ‘simple’ educational tasks, such as single-digit addition for young kids.” 

“In order for the machine teaching approach to work, it needs a good model of how the learner behaves—that is, how the learner’s behavior changes with different kinds of learning or practice experiences,” said Rogers. “Also, the model needs to be computational; it has to be able to make concrete, quantitative predictions about the learner’s behavior.”

According to Zhu, machine teaching will work off two inputs, one being the individual student’s machine learning algorithm in their head, and two the educational goal. Machine teaching then constructs the optimal lesson for a student to achieve the educational goal, he says.

The hope is to use the work to assist teachers in developing lesson plans and curricula across a wide range of disciplines.


The highly-anticipated educational tracks for the 2015 R&D 100 Awards & Technology Conference feature 28 sessions, plus keynote speakers Dean Kamen and Oak Ridge National Laboratory Director Thom Mason.  Learn more.

Source: The Flipside of Machine Learning

Via: Google Alert for ML