Learning science and Machine Learning intersect in the personalized learning layers that Cerego is powering for parties ranging from EdX MOOCs, to Arizona State University’s Global Freshman Academy courses, to publishers such as Elsevier and McGraw-Hill Education.
The goal for Cerego-enhanced online learning is to help users capture knowledge – and to do it faster and hold on to it longer. Andrew Smith Lewis, Founder and Executive Chairman, says the company’s genesis lies in taking applied research from fields including neuroscience, cognitive science, and experimental psychology, and acting on it in the service of the science of memory.
“Our desire is to help people learn and retain knowledge and know where they are on a personalized level,” says Smith Lewis. With Cerego, course creators provide the actual content, indicating to the system the critical knowledge they want their students to take away. The system analyzes it and determines what questions to ask to reflect that, and in what format to ask them. The other intellectual property involved in Cerego is that it can look at the individual learner and determine the right timing to present content to that person and in what sequence to present it.
Research Meets Technology
Cerego puts core principles of validated learning science research to work via technology that includes a proprietary algorithm, the company explains. It assesses students’ individual memory strengths based on their performance; the forgetting curve that indicates how human memory fades over time; and, distributed practice formulas. “The algorithm is capable of taking content and figuring out the optimal time [for an individual] to review it in future in order to promote long-term memory,” Smith Lewis says.
The error-correcting adaptive controller that is a key part of its Machine Learning foundation is meant to ensure that Cerego constantly learns as it travels through each unique user’s mastery-of-content path, item by item. Applying its error-correcting model to memory and learning engines is unique, Smith Lewis says.
In practice, on the first go-round, Cerego teaches the content to users based on what the instructor/content provider determines the student needs to learn. It follows that up with the review mode where it generates questions about the content on its own. “That is part of the Machine Learning in that the content author puts in the structured knowledge and the system creates questions that map to it on the fly,” Smith Lewis notes. The instructor or publisher is removed from the process of having to create those queries himself or herself.
Based on the student’s review interactions, Cerego determines how well the individual knows the information at that point and when the best time is to test retention. Cerego’s technology then matches how students perform against what its performance expectations were. “It plugs the gap between the two into its calculations on an item by item basis so that the next time it makes a prediction it becomes more accurate,” Smith Lewis says.
Profiles of individual students’ learning progress on an individual item basis are informed by factors such as the type of queries presented (diagram vs. multiple choice, for instance), and whether or not a student answers a question correctly the first time it is asked or after the student goes through the exercise again. Whether or not the content is classified as difficult or easy or moderate is based on the student’s response, rather than some objective criteria. Cerego’s Memory Bank for a student is updated as the learner demonstrates mastery of one item, so that it won’t continue to barrage the student with that topic for a period of time.
Similarly, the Memory Bank is updated if the student shows that he or she has had trouble responding correctly to a different topic, so that instruction on it will be presented again the next time the individual logs in. “It granularly pinpoints where you are and where you need to move to,” Smith Lewis says.
Parts of students’ brains are always calculating what they need to do: what course materials to review tonight, what tests to prep for on the weekend, and so on. “We are always making meta-cognitive decisions about what we need to do,” says Smith Lewis. “Cerego supplants that for us.”
Any Topic, Any Time
Cerego is platform independent. The ability to run on mobile devices for anytime, anywhere learning is important, he explains. “The recipe for long-term lasting memory is little and often, so mobile is huge for us,” he says.
The technology also is content agnostic, so that it can work with a wide range of partners to help their students on “the how” of learning and remembering. Whether an individual is studying astronomy, music, or veterinary science, it drives at facilitating durable knowledge and giving insight into current and future performance.
With its partner Elsevier, for instance, Cerego has ported the health publisher’s nursing textbooks to its platform. Students using Elsevier’s nursing portal, Evolve, can see the textbooks for which they are provisioned, and head over to the orthopedic surgery chapter, for instance. From there, they can fire up the Cerego learning experience and discover that they have seven fading memories recommended for review against the visualized backdrop of their individual Memory Banks, which show the constellation of items they are studying in a particular unit and the content for which they will be responsible over the coming months.
While nursing students may be reviewing their knowledge of fibulas, individuals enrolled in the Jazz music appreciation MOOC from the University of Texas at Austin on EdX can be sharpening their listening comprehension or key facts about that genre. MOOCs tend to have a lower completion rate – around 5%, Smith Lewis says, but “some MOOCs powered by Cerego end up with a 12 to 13% completion rate. The app can drive engagement and retention on some of these platforms.”
Cerego also has entered into a deal with school textbook publisher VitalSource to work with its e-textbook platform, VitalSource Bookshelf, and expects to have an initial product there this quarter that marries reading experiences to Cerego’s smarts. “A digital textbook is all fine and well, but the next step of that is adding a type of learning intelligence and analytics to the book,” says Smith Lewis.
Instructors always can make their instruction content as elaborate and visual as they want, he explains. On their side of the platform, instructors can log in and view a Memory Bank that shows a constellation of their students and what levels these learners are at, when they last studied, how the content appears to students – difficult or moderate, for instance – based on analytics, and what their study time for content amounts to.
“It’s real-time analytics about where they are at any moment,” says Smith Lewis. Instructors can use content analytics information to see, for instance, what is the percentage rate of correct answers for particular topics for students. If it’s low, it’s an opportunity to do an intervention in that area before moving forward with the class (should the learning experience be tied to a live class or online courses with discussion board or group chat components). “It’s a huge advantage in data-driven decisions,” he says.
Next Steps for Cerego
Cerego sees its platform also having a role in the corporate and professional world. It’s launched a project with a multinational logistics company that’s using it for new driver safety training. “New employees have lots of stuff to learn, and wouldn’t it be cool if that was quantifiable,” Smith Lewis says. “That’s a big push for us.”
Smith Lewis thinks Cerego’s technology could also be an asset in many other fields where employees need a strong grasp in various aspects of foundational knowledge. The health and aeronautics sectors are among them. Cerego won’t make you a proficient pilot alone, he acknowledges, “but if you need to know the instrumentation ratings for a certain aircraft it’s a great way to take content with you and use adaptive algorithms to ensure you know it,” he says.
Beyond its focus on the education and business markets, though, Cerego also lets consumers have free and open access to its platform to create their own learning content and collaborate with others on it, too. The company also is aiming to push the envelope of what its Machine Learning algorithm can do. While it’s been able to apply its technology to utilize the wealth of research that exists on long-term memory and retention, Smith Lewis sees the chance to perhaps even further personalize its algorithm to put it to work for “intelligent cramming,” as well.