CMU Commits to Continue LearnLab as part of the Simon Initiative
Motivated to a large degree on creating a sustainable home for LearnLab, one of the first things Dr. Subra Suresh did when he arrived at CMU as its new President in 2013 was start the Simon Initiative (see lower left Figure 6) and the associated Global Learning Initiative. The Simon Initiative harnesses a cross-disciplinary learning engineering ecosystem that has developed over several decades at Carnegie Mellon and was particularly influenced by LearnLab, the associated Open Learning Initiative, and by CMU’s Eberly Teaching Center. The Simon Initiative has been providing seed grants to researchers to start new learning science research and to employ learning science and technology in course improvement. It is also funding DataLab, a broadening of DataShop to not only support learning scientists, but also course instructors and developers in using data to improve learning. LearnLab will continue as the scientific arm of the Simon Initiative.
Masters of Educational Technology and Applied Learning Science (METALS)
A central element of sustainability of the LearnLab education strategy was the creation, in 2013, of the Masters of Educational Technology and Applied Learning Science (METALS; see lower left Figure 6). The curriculum is an outgrowth of the extensive research conducted by LearnLab.
METALS is a one-year, interdisciplinary masters program that trains graduate students to apply evidence-based research in learning to create effective instruction and educational technologies within formal and informal settings such as schools, workplaces, and museums. The professional program culminates with a seven-month capstone project for an external industry client. At its core, the METALS program is a union of the Human-Computer Interaction Institute and the Department of Psychology, but it draws from cognitive science, statistics, computer science, education and design. To understand how people learn, our students are trained in applying cognitive science methods to map learning objectives to sub-objectives and finally to well-designed instructional activities. We need to collect and analyze student progress. To do that our students learn how to use statistics to analyze extremely large data sets. These analyses drive continuous improvement. To create effective online educational courseware, our students need to understand at least the basics of tutor creation. We teach our students to use software engineering techniques to create these tutors. We teach our students how to design an appropriate curriculum using the latest theories.
New NSF Funding to Expand DataShop into LearnSphere
We are sustaining and extending DataShop with the help of a $5 million grant from NSF Cyberinfrastructure (grant #ACI-1443068) that will create LearnSphere (see learnsphere.org). This project is the first Cyberinfrastructure grant dedicated to education and involves a collaboration with CMU, MIT, Stanford, and University of Memphis. LearnSphere will extend DataShop’s focus on clickstream data from educational technology interactions to other kinds of educational data including online course (MOOC) data (e.g., lecture video watching), discussion board data, and other forms of dialogue data. It builds off of technologies developed in the LearnLab SC thrust. LearnSphere will provide a distributed data infrastructure (e.g., we already have a DataShop instance at Memphis) to support analytic methods that integrate across a wide range of learning data types so as to produce discoveries not possible within current data silos.
Created a New Field: Educational Data Mining
The field of Educational Data Mining (EDM) was founded under the primary leadership of Ryan Baker while he was a postdoc at LearnLab and the first technical director of DataShop, but also with active involvement of other LearnLab contributors including John Stamper, Carolyn Rose, Ken Koedinger, Kurt VanLehn, Brian Junker, Steve Ritter, Geoff Gordon, Vincent Aleven, Bruce McLaren, Noboru Matsuda and others. The field is concerned with developing methods for exploring the unique and increasingly large-scale data that come from educational settings, and using those methods to better understand students, and the settings in which they learn. These are goals that have been central to LearnLab from the beginning.
A key to the creation of the field was a number of EDM workshops at related conferences (AIED 2005 & 7, EC-TEL 2007, ICALT 2007, UM 2007, AAAI 2005, ITS 2000, 4, & 6) that culminated in the formation of an annual International Conference on Educational Data Mining in 2008. LearnLab researchers were central organizers and participants in these early workshops and in all of the conferences since. LearnLab produced the best paper at the first conference (Shih, Koedinger, & Scheines, 2008) for work demonstrating an unobtrusive detector of a student self-regulated learning behavior called self explanation. Each year, LearnLab has contributed a large fraction of the papers at the conference and has continued to play a leading role at the conference and in the field:
- 2008: Best paper: A response time model for bottom-out hints as worked examples (Shih, Koedinger, & Scheines)
- 2009: Best paper: Reducing the Knowledge Tracing Space (Ritter, Harris, Nixon, Dickison, Murray & Towle)
- 2010: EDM Conference host in Pittsburgh; Organized the 2010 KDD Cup, the first ever educational data mining competition at the Knowledge Discovery and Data Mining conference. Organized by LearnLab leaders John Stamper, Ken Koedinger, & Geoff Gordon.
- 2011: Honoring his crucial role in forming the EDM field, Ryan Baker (former DataShop Director) was elected as the inaugural president of the International Educational Data Mining Society. New DataShop Director John Stamper was an invited speaker at the EDM conference.
- 2012: Best paper: Automated Student Model Improvement (Koedinger, McLaughlin, & Stamper). Two Other Best Paper Nominees: Learner Differences in Hint Processing (Goldin, Koedinger, & ALeven); The Impact on Individualizing Student Models on Necessary Practice Opportunities (Lee & Brunskill). Best student paper: Comparison of methods to trace multiple subskills: Is LR-DBN best? (Xu & Mostow)
- 2013: Best paper: Does Representational Understanding Enhance Fluency – Or Vice Versa? Searching for Mediation Models (Rau, Scheines, Aleven, Rummel); Two Other Best Paper Nominees: Predicting Player Moves in an Educational Game: A Hybrid Approach (Liu, Mandel, Butler, Andersen, O’Rourke, Brunskill, Popovic); Applying three models of learning to individual student log data (van de Sande); Best student paper: A spectral learning approach to knowledge tracing. (Falakmasir, Pardos, Gordon & Brusilovsky) and nominee for best student paper: 55: Discovering Student Models with a Clustering Algorithm Using Problem Content (Li, Cohen, Koedinger)
- 2014: Program chair: John Stamper; Conference chair: Bruce McLaren
Ryan Baker created and taught a free MOOC on educational data mining entitled “Big Data in Education” via Coursera in October 2013. It remains online at http://www.columbia.edu/~rsb2162/bigdataeducation.html. He also started a masters of science in learning analytics at Teachers College, Columbia University. Since the founding of this field, Northeastern also offers a similar degree and others like Brandeis University and Algonquin College offer certificates in learning analytics. We expect this growth to continue as we see more programs develop around educational data mining and learning analytics.