Connecting Talent to Education at Massive Scale

Our open-to-the-public LinkedIn tech talks have always been exciting to people interested in big datamachine learning or distributed systems. However, the most popular event in the series highlighted a different type of scalable online learning -- one where  humans are interpreting new data, receiving feedback, and constantly improving.

Our speakers were Daphne Koller and Andrew Ng, world-renowned computer science professors at Stanford and machine learning superstars. They recently launched Coursera, which is partnering with Princeton, Stanford, the University of Michigan, and the University of Pennsylvania to offer free online courses to everyone.

Their mission is inspiring, as are the technological challenges they are solving in order to accomplish it. As computer scientists, our ambition is to achieve scalability through technology. Until recently, however, education has resisted this widespread trend towards greater efficiency. College tuition costs have been rising twice as quickly as healthcare costs -- and four times more quickly than the Consumer Price Index (CPI). And yet education is essential to filling the widening talent gap: in the United States, we have the highest number of job openings in nearly four years, while 22 million people are unemployed or underemployed.

Coursera brings technology (and scale) to education. Over a million students have signed up, and some courses reach tens of thousands --  numbers unimaginable in a traditional classroom setting. Andrew noted that it would have taken him 250 years to teach as many students as he reached with his online machine learning class.

Is it possible to achieve that scale while providing the personalized instruction that studies show is best for learning? Coursera starts by using world-class instructors, then goes beyond passive video instruction by offering in-lecture interactive quizzes -- the equivalent of professors asking quick questions in class. However, in contrast to what amounts to semi-rhetorical questions in a large auditorium setting, the quizzes are in sync with every student and provide immediate feedback and recall before a student has the chance to fall behind.

Coursera’s approach to feedback and assessment is a very interesting application of data science. Tests are either computer-graded or peer-graded -- the latter following industry tested crowdsourcing best practices (clear instructions, gold standards, training, qualification tasks assessor agreement monitoring etc.). Peer grading isn’t just treated as a means for scaling -- it is part of the learning process. One of Daphne’s charts showed that students significantly improved on subsequent tests after peer- and self-grading. Interestingly, the better students learned even more from self-grading than from grading others.

Coursera’s scale also creates a powerful community. Students who ask questions in online forums receive answers in 20 minutes, and voting mechanisms surface top questions. Study groups meet all over the world to meet students’ particular needs -- be it language, learning style or background. Employers are also taking notice -- we are starting to see Andrew’s machine learning class on the resumes and LinkedIn profiles of people we’re trying to hire.

Of course, scale also offers opportunities for data analytics. Instructors can detect patterns in students’ mistakes, common learning paths, and particularly challenging modules. They can then adapt their material based on these insights. In a particular apropos application of machine learning, Coursera identifies forum answers that are most predictive of submitting a correct homework assignment after submitting one that failed the automated verification tests.

It was a privilege to have Daphne and Andrew speak about their newest endeavor here at LinkedIn. Our mission is to connect talent to opportunity at massive scale, and we cannot imagine a better complement to that mission than revolutionizing education.