MOOC Learning Analytics
Description: The advent of MOOCs has created a great opportunity for research in the area of learning science, because of its massive adoption (in the order of millions of students) and the digital nature of the platforms. In this project, we analyse data generated be the Coursera and EdX platforms to understand how people learn and which learning strategies lead to better (or worst) results. The data we use for this analysis includes: MOOC grades, problem submissions, video navigation patterns, and forum interactions.
Flipped Classroom Models
Partners: CEDE, EPFL Professors
Description: The availability of MOOCs as a choice for teaching university courses has been useful not only to students, but also to teachers. Some of the teachers at EPFL who authored a MOOC have experimented with a concept called “flipped classroom”, for structuring their EPFL courses. The model consists of asking students to watch video lectures at home, and then reserve the class for use-case discussions (see example). The goal of this project is to analyse the impact of MOOC usage on the grades of students at EPFL. This requires comparing our students’ MOOC learning patterns and grades with their EPFL grades in the respective courses.
Description: The Learning Companion is a digital assistant for EPFL students that we’re developing in collaboration with CAPE and the PocketCampus team (available here). This app allows students to track their own study habits, and professors to monitor their progress in class. It then provides useful analytics and feedback based on the latest research in learning science. The goal is to help students become more self-aware of their learning habits, and follow the best practices proposed by experts in the field. For this purpose, we collect self-reported data from students, such as the types of difficulties they had solving assignments, and questionnaires that measure their methods of learning. None of this data is used for course evaluation purposes.
Description: The EPFL course catalog is one of the most interesting sources of data at EPFL that remains largely untapped. Using machine learning techniques, such as natural language processing, we can analyse courses’ content and understand how they are related to each other—in particular, which courses are pre-requisites to others. This is a new way of navigating the course catalog that students can use to better plan their studies at EPFL (see demo). We can take this goal further and actually estimate dependencies between courses based on past student choices and performance (e.g., by looking at their grades in specific combinations of courses).
Performance-Based Curriculum Optimisation
Description: The course curriculum at EPFL is structured every year based on a variety of criteria, such as their relevance for the intellectual foundation that students are supposed to develop, but also the trends in the scientific world, the needs of the job market, etc. We propose adding a new indicator to the equation, based on students’ sequential performance in courses. The idea is to analyse student grades, and understand which combinations of courses produced the best results. This will help policy makers implement better informed changes to the curriculum, based on a detailed analysis of student performance.
Course Recommendation System
Description: One of the features that students would love to see the most in course descriptions, based on informal conversations we had with students at EPFL, is a course recommender system. In the areas of machine learning and artificial intelligence, recommender systems are a hot topic and EPFL labs have no lack of knowhow in this area. We have been working on a course recommendation algorithm, based on artificial neural networks, that uses past student choices to predict their choices in the future. This requires us to analyse which courses students took in which semesters.
Academic Concepts Database
Description: One of the goals of courses in a university-level STEM field is teach students different concepts, and how they relate to each other to generate natural phenomena or to solve a specific technical problem. If we “atomise” these courses, by breaking them down to the smallest self-contained concepts, and then represent them as a dependency graph, we obtain a valuable new tool for visualising the EPFL course catalog, as well as planning future modifications. A concepts graph would allow us, for instance, to connect subjects taught at EPFL to skills required by the job market.
Job Market Skills Graph
Description: One of the core missions of EPFL is to prepare the young population for the Swiss job market, not only by providing a solid university education but also by providing a breadth of relevant courses in the continuous education departement (including MOOCs). To help us understand the real needs of the job market, modern machine learning can be a powerful tool. Using natural language processing techniques, combined with graph representation models, we can obtain a new and revealing perspective of the hard and soft skills that are in highest demand (see demo).
Data-Based Analysis of Students’ Career Choices
Description: The path of students who come to EPFL for their studies does not end when they graduate. As one of the best universities in the world, we are committed to providing the best future to our Alumni, particularly in terms of opportunities they get in the job market. It is in the best interests of both parties (student and university) to make sure our graduates are highly skilled in relevant areas, and therefore in high demand. Yet, very little has been done at EPFL to follow up on our Alumni’s path after graduation. We intend to change that, and begin putting our labs’ knowhow into use—particularly in the areas of data science and machine learning—for the purpose of career planning. This is a long-term project that will involve building a database of Alumni’s career choices, and connecting these choices back to the concepts they’ve learned at EPFL.
Data-Based Analysis of Graduate Students’ Migration
Description: A thorough empirically-based economic analysis can go beyond an overview of foreign Swiss-educated graduate students’ retention rates and gauge the selection process behind their outmigration. Are there differences between who stays and who leaves in terms of observable quality? Which outmigration determinants can be identified? Focusing on STEM taught- and research-oriented graduates from the Swiss Federal Institute of Technology in Lausanne, the proposed research aims at measuring Swiss international students’ outmigration selectivity. Through a reduced-form econometric approach, we will be able to characterise outmigration selection as proxied by observable characteristics such as coursework performance (e.g., GPA for Master’s students), and publication records (e.g., citations received, journals’ impact factor for PhD students).
Data-Based Analysis of Tech-Transfer at EPFL
Description: As an alternative to a career in academia or the corporate world, many students who graduate from EPFL have the ambition of creating their own startup, particularly upon completing their Ph.D. programs. Unfortunately, neither students nor the administration have access to precise data analytics concerning tech-transfer at EPFL, with indicators such as: hot spots of VC investment, research areas with high tech-transfer ROI, and coursework that stimulates entrepreneurship among students. The goal of this project is to tap on existing data (including courses students took, papers they published, and startups they’ve created) and produce meaningful analytics for students and the administration.
Research Collaborations Graph
Description: Research productivity is one of the best indicators of a successful research university like MIT, Stanford, and EPFL. In order to be productive, researchers need to publish… a lot! This is where collaborations come in. One of the most efficient ways for a researcher to increase its publications output is to collaborate with other researchers. Many times, however, people at EPFL (and in the scientific community at large) don’t know what kind of work each other is doing. Some people are working on similar or related problems, and, not knowing about it, could potentially collaborate and publish papers together. The goal of this project is to analyse the EPFL publications graph, and estimate potential collaborations by strategically connecting nodes in the graph.
“DataJam Days” Hackaton
Description: To promote a culture of collaboration around data sharing and reuse, the Swiss Data Science Center (SDSC) organises the Data Jam Days—an event for anyone interested in analysing, reusing, exploring, and visualising scientific data. Open data can lower some of the barriers to productive collaborations. As science is struggling with a reproducibility crisis, EPFL wants to promote a culture of transparency in which researchers openly publish research reports together with the underlying data to allow scrutiny and reuse by others. The datasets we provided for this hackaton were anonymised.
ArtLab Datasquare Monolith Project
Description: The Datasquare monolith is a giant interactive dark screen (Star Trek style) that displays a variety of EPFL facts and figures in an artful, user-friendly way. It includes mappings of the EPFL ecosystem, student populations, areas of research, publication graphs, and MOOCs patterns of activity. The CEDE collaborated with the ArtLab team to develop the MOOCs visualisation functionality of the monolith (see demo).