07-31, 12:30–13:00 (UTC), Red Track
Data science and machine learning courses are in high demand with growing enrollments. We discuss our experience teaching a computational DS&ML course with 250+ students that is designed to scale. We highlight the use of experiential learning, just-in-time presentation of key concepts, and real-time feedback on students’ understanding. We demo a computational textbook to make teaching a scalable DS&ML course easy.
DEMO: https://mynerva.io/courses/register, code juliacon2020
Over the past two years at the University of Michigan, we have successfully taught computational data science and machine learning using Julia to over a thousand graduate students, spanning over 80 disciplines. We've scaled to support 275 enrolled students (Fall 2019) and expect to scale to over 500 in the very near future. The course is taught in a hybrid lecture-lab style which presents students with the opportunity to experience hands-on, interactive, inquiry-based learning. Student feedback has been overwhelmingly positive. We will present a hands-on demo, featuring selected content from the course, and invite the audience to experience it for themselves.
We will present our vision for an interactive computational textbook platform designed to stimulate deeper conversations between students and the material that they are attempting to master. The primary technique used to facilitate this deeper engagement is the punctuation of instructional content (such as motivation, math fundamentals, theorems, proofs, etc.) with short, self-contained exercises that are designed to allow the student to constantly check their understanding. Exercises include conceptual questions to check comprehension of the underlying content, auto-graded programming assignments to practice the ability to put ideas into practice as code, and free-response prompts to allow the student to make their own observations and draw the appropriate conclusions.
The effect of this constant "self-checking" is two-fold. First, students are forced to more rigorously develop foundational knowledge before moving on to more advanced topics. This sets students up for success when the more advanced content is presented and avoids the dreaded "content paralysis" phenomenon where students are overwhelmed by the sheer scope of content on first glance (this usually manifests itself as a student blindly running code in “run-all mode" in an attempt to produce the desired results instead of actually understanding what they are doing). Second, this framework allows an author to more carefully design content to encourage a guided learning experience. The author can design a narrative, complete with exposition (what is this algorithm?), conflict (when does it fail?), and resolution (how can we fix it?). Our demo will allow you to experience how much more immersive “living” computational textbooks written in this way can be.
Travis DePrato is an Experiential Learning Software Developer at the University of Michigan, working on developing new ways to teach courses in computer science, computational mathematics, and data science.
Professor of EECS at University of Michigan